Jump to content
documentation.suse.com / System Analysis and Tuning Guide
SUSE Linux Enterprise Server 12 SP5

System Analysis and Tuning Guide

An administrator's guide for problem detection, resolution and optimization. Find how to inspect and optimize your system by means of monitoring tools and how to efficiently manage resources. Also contains an overview of common problems and solutions and of additional help and documentation resources.

Publication Date: November 28, 2024

Copyright © 2006–2024 SUSE LLC and contributors. All rights reserved.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or (at your option) version 1.3; with the Invariant Section being this copyright notice and license. A copy of the license version 1.2 is included in the section entitled GNU Free Documentation License.

For SUSE trademarks, see https://www.suse.com/company/legal/. All third-party trademarks are the property of their respective owners. Trademark symbols (®, ™ etc.) denote trademarks of SUSE and its affiliates. Asterisks (*) denote third-party trademarks.

All information found in this book has been compiled with utmost attention to detail. However, this does not guarantee complete accuracy. Neither SUSE LLC, its affiliates, the authors nor the translators shall be held liable for possible errors or the consequences thereof.

About This Guide

SUSE Linux Enterprise Server is used for a broad range of usage scenarios in enterprise and scientific data centers. SUSE has ensured SUSE Linux Enterprise Server is set up in a way that it accommodates different operation purposes with optimal performance. However, SUSE Linux Enterprise Server must meet very different demands when employed on a number crunching server compared to a file server, for example.

It is not possible to ship a distribution that is optimized for all workloads. Different workloads vary substantially in some aspects. Most important among those are I/O access patterns, memory access patterns, and process scheduling. A behavior that perfectly suits a certain workload might reduce performance of another workload. For example, I/O-intensive tasks, such as handling database requests, usually have completely different requirements than CPU-intensive tasks, such as video encoding. The versatility of Linux makes it possible to configure your system in a way that it brings out the best in each usage scenario.

This manual introduces you to means to monitor and analyze your system. It describes methods to manage system resources and to tune your system. This guide does not offer recipes for special scenarios, because each server has got its own different demands. It rather enables you to thoroughly analyze your servers and make the most out of them.

Part I, “Basics”

Tuning a system requires a carefully planned proceeding. Learn which steps are necessary to successfully improve your system.

Part II, “System Monitoring”

Linux offers a large variety of tools to monitor almost every aspect of the system. Learn how to use these utilities and how to read and analyze the system log files.

Part III, “Kernel Monitoring”

The Linux kernel itself offers means to examine every nut, bolt and screw of the system. This part introduces you to SystemTap, a scripting language for writing kernel modules that can be used to analyze and filter data. Collect debugging information and find bottlenecks by using kernel probes and Perf. Last, monitor applications with Oprofile.

Part IV, “Resource Management”

Learn how to set up a tailor-made system fitting exactly the server's need. Get to know how to use power management while at the same time keeping the performance of a system at a level that matches the current requirements.

Part V, “Kernel Tuning”

The Linux kernel can be optimized either by using sysctl, via the /proc and /sys file systems or by kernel command line parameters. This part covers tuning the I/O performance and optimizing the way how Linux schedules processes. It also describes basic principles of memory management and shows how memory management can be fine-tuned to suit needs of specific applications and usage patterns. Furthermore, it describes how to optimize network performance.

Part VI, “Handling System Dumps”

This part enables you to analyze and handle application or system crashes. It introduces tracing tools such as strace or ltrace and describes how to handle system crashes using Kexec and Kdump.

Tip
Tip: Getting the SUSE Linux Enterprise SDK

The SDK is a module for SUSE Linux Enterprise and is available via an online channel from the SUSE Customer Center. Alternatively, go to http://download.suse.com/, search for SUSE Linux Enterprise Software Development Kit and download it from there. Refer to Book “Deployment Guide”, Chapter 15 “Installing Modules, Extensions, and Third Party Add-On Products” for details.

1 Available documentation

Online documentation

Our documentation is available online at https://documentation.suse.com. Browse or download the documentation in various formats.

Note
Note: Latest updates

The latest updates are usually available in the English-language version of this documentation.

SUSE Knowledgebase

If you have run into an issue, also check out the Technical Information Documents (TIDs) that are available online at https://www.suse.com/support/kb/. Search the SUSE Knowledgebase for known solutions driven by customer need.

Release notes

For release notes, see https://www.suse.com/releasenotes/.

In your system

For offline use, the release notes are also available under /usr/share/doc/release-notes on your system. The documentation for individual packages is available at /usr/share/doc/packages.

Many commands are also described in their manual pages. To view them, run man, followed by a specific command name. If the man command is not installed on your system, install it with sudo zypper install man.

2 Improving the documentation

Your feedback and contributions to this documentation are welcome. The following channels for giving feedback are available:

Service requests and support

For services and support options available for your product, see https://www.suse.com/support/.

To open a service request, you need a SUSE subscription registered at SUSE Customer Center. Go to https://scc.suse.com/support/requests, log in, and click Create New.

Bug reports

Report issues with the documentation at https://bugzilla.suse.com/.

To simplify this process, click the Report an issue icon next to a headline in the HTML version of this document. This preselects the right product and category in Bugzilla and adds a link to the current section. You can start typing your bug report right away.

A Bugzilla account is required.

Contributions

To contribute to this documentation, click the Edit source document icon next to a headline in the HTML version of this document. This will take you to the source code on GitHub, where you can open a pull request.

A GitHub account is required.

Note
Note: Edit source document only available for English

The Edit source document icons are only available for the English version of each document. For all other languages, use the Report an issue icons instead.

For more information about the documentation environment used for this documentation, see the repository's README.

Mail

You can also report errors and send feedback concerning the documentation to <>. Include the document title, the product version, and the publication date of the document. Additionally, include the relevant section number and title (or provide the URL) and provide a concise description of the problem.

3 Documentation conventions

The following notices and typographic conventions are used in this document:

  • /etc/passwd: Directory names and file names

  • PLACEHOLDER: Replace PLACEHOLDER with the actual value

  • PATH: An environment variable

  • ls, --help: Commands, options, and parameters

  • user: The name of a user or group

  • package_name: The name of a software package

  • Alt, AltF1: A key to press or a key combination. Keys are shown in uppercase as on a keyboard.

  • File, File › Save As: menu items, buttons

  • AMD/Intel This paragraph is only relevant for the AMD64/Intel 64 architectures. The arrows mark the beginning and the end of the text block.

    IBM Z, POWER This paragraph is only relevant for the architectures IBM Z and POWER. The arrows mark the beginning and the end of the text block.

  • Chapter 1, Example chapter: A cross-reference to another chapter in this guide.

  • Commands that must be run with root privileges. You can also prefix these commands with the sudo command to run them as a non-privileged user:

    root # command
    tux > sudo command
  • Commands that can be run by non-privileged users:

    tux > command
  • Commands can be split into two or multiple lines by a backslash character (\) at the end of a line. The backslash informs the shell that the command invocation will continue after the line's end:

    tux > echo a b \
    c d
  • A code block that shows both the command (preceded by a prompt) and the respective output returned by the shell:

    tux > command
    output
  • Notices

    Warning
    Warning: Warning notice

    Vital information you must be aware of before proceeding. Warns you about security issues, potential loss of data, damage to hardware, or physical hazards.

    Important
    Important: Important notice

    Important information you should be aware of before proceeding.

    Note
    Note: Note notice

    Additional information, for example about differences in software versions.

    Tip
    Tip: Tip notice

    Helpful information, like a guideline or a piece of practical advice.

  • Compact Notices

    Note

    Additional information, for example about differences in software versions.

    Tip

    Helpful information, like a guideline or a piece of practical advice.

4 Support

Find the support statement for SUSE Linux Enterprise Server and general information about technology previews below. For details about the product lifecycle, see https://www.suse.com/lifecycle.

If you are entitled to support, find details on how to collect information for a support ticket at https://documentation.suse.com/sles-15/html/SLES-all/cha-adm-support.html.

4.1 Support statement for SUSE Linux Enterprise Server

To receive support, you need an appropriate subscription with SUSE. To view the specific support offers available to you, go to https://www.suse.com/support/ and select your product.

The support levels are defined as follows:

L1

Problem determination, which means technical support designed to provide compatibility information, usage support, ongoing maintenance, information gathering and basic troubleshooting using available documentation.

L2

Problem isolation, which means technical support designed to analyze data, reproduce customer problems, isolate a problem area and provide a resolution for problems not resolved by Level 1 or prepare for Level 3.

L3

Problem resolution, which means technical support designed to resolve problems by engaging engineering to resolve product defects which have been identified by Level 2 Support.

For contracted customers and partners, SUSE Linux Enterprise Server is delivered with L3 support for all packages, except for the following:

  • Technology previews.

  • Sound, graphics, fonts, and artwork.

  • Packages that require an additional customer contract.

  • Some packages shipped as part of the module Workstation Extension are L2-supported only.

  • Packages with names ending in -devel (containing header files and similar developer resources) will only be supported together with their main packages.

SUSE will only support the usage of original packages. That is, packages that are unchanged and not recompiled.

4.2 Technology previews

Technology previews are packages, stacks, or features delivered by SUSE to provide glimpses into upcoming innovations. Technology previews are included for your convenience to give you a chance to test new technologies within your environment. We would appreciate your feedback. If you test a technology preview, please contact your SUSE representative and let them know about your experience and use cases. Your input is helpful for future development.

Technology previews have the following limitations:

  • Technology previews are still in development. Therefore, they may be functionally incomplete, unstable, or otherwise not suitable for production use.

  • Technology previews are not supported.

  • Technology previews may only be available for specific hardware architectures.

  • Details and functionality of technology previews are subject to change. As a result, upgrading to subsequent releases of a technology preview may be impossible and require a fresh installation.

  • SUSE may discover that a preview does not meet customer or market needs, or does not comply with enterprise standards. Technology previews can be removed from a product at any time. SUSE does not commit to providing a supported version of such technologies in the future.

For an overview of technology previews shipped with your product, see the release notes at https://www.suse.com/releasenotes.

Part I Basics

  • 1 General Notes on System Tuning
  • This manual discusses how to find the reasons for performance problems and provides means to solve these problems. Before you start tuning your system, you should make sure you have ruled out common problems and have found the cause for the problem. You should also have a detailed plan on how to tune the system, because applying random tuning tips often will not help and could make things worse.

1 General Notes on System Tuning

This manual discusses how to find the reasons for performance problems and provides means to solve these problems. Before you start tuning your system, you should make sure you have ruled out common problems and have found the cause for the problem. You should also have a detailed plan on how to tune the system, because applying random tuning tips often will not help and could make things worse.

Procedure 1.1: General Approach When Tuning a System
  1. Specify the problem that needs to be solved.

  2. In case the degradation is new, identify any recent changes to the system.

  3. Identify why the issue is considered a performance problem.

  4. Specify a metric that can be used to analyze performance. This metric could for example be latency, throughput, the maximum number of users that are simultaneously logged in, or the maximum number of active users.

  5. Measure current performance using the metric from the previous step.

  6. Identify the subsystem(s) where the application is spending the most time.

    1. Monitor the system and/or the application.

    2. Analyze the data, categorize where time is being spent.

  7. Tune the subsystem identified in the previous step.

  8. Remeasure the current performance without monitoring using the same metric as before.

  9. If performance is still not acceptable, start over with Step 3.

1.1 Be Sure What Problem to Solve

Before starting to tuning a system, try to describe the problem as exactly as possible. A statement like The system is slow! is not a helpful problem description. For example, it could make a difference whether the system speed needs to be improved in general or only at peak times.

Furthermore, make sure you can apply a measurement to your problem, otherwise you cannot verify if the tuning was a success or not. You should always be able to compare before and after. Which metrics to use depends on the scenario or application you are looking into. Relevant Web server metrics, for example, could be expressed in terms of:

Latency

The time to deliver a page

Throughput

Number of pages served per second or megabytes transferred per second

Active Users

The maximum number of users that can be downloading pages while still receiving pages within an acceptable latency

1.2 Rule Out Common Problems

A performance problem often is caused by network or hardware problems, bugs, or configuration issues. Make sure to rule out problems such as the ones listed below before attempting to tune your system:

  • Check the output of the systemd journal (see Book “Administration Guide”, Chapter 16 “journalctl: Query the systemd Journal”) for unusual entries.

  • Check (using top or ps) whether a certain process misbehaves by eating up unusual amounts of CPU time or memory.

  • Check for network problems by inspecting /proc/net/dev.

  • In case of I/O problems with physical disks, make sure it is not caused by hardware problems (check the disk with the smartmontools) or by a full disk.

  • Ensure that background jobs are scheduled to be carried out in times the server load is low. Those jobs should also run with low priority (set via nice).

  • If the machine runs several services using the same resources, consider moving services to another server.

  • Last, make sure your software is up-to-date.

1.3 Finding the Bottleneck

Finding the bottleneck very often is the hardest part when tuning a system. SUSE Linux Enterprise Server offers many tools to help you with this task. See Part II, “System Monitoring” for detailed information on general system monitoring applications and log file analysis. If the problem requires a long-time in-depth analysis, the Linux kernel offers means to perform such analysis. See Part III, “Kernel Monitoring” for coverage.

Once you have collected the data, it needs to be analyzed. First, inspect if the server's hardware (memory, CPU, bus) and its I/O capacities (disk, network) are sufficient. If these basic conditions are met, the system might benefit from tuning.

1.4 Step-by-step Tuning

Make sure to carefully plan the tuning itself. It is of vital importance to only do one step at a time. Only by doing so can you measure if the change provided an improvement or even had a negative impact. Each tuning activity should be measured over a sufficient time period to ensure you can do an analysis based on significant data. If you cannot measure a positive effect, do not make the change permanent. Chances are, that it might have a negative effect in the future.

Part II System Monitoring

  • 2 System Monitoring Utilities
  • There are number of programs, tools, and utilities which you can use to examine the status of your system. This chapter introduces some and describes their most important and frequently used parameters.

  • 3 Analyzing and Managing System Log Files
  • System log file analysis is one of the most important tasks when analyzing the system. In fact, looking at the system log files should be the first thing to do when maintaining or troubleshooting a system. SUSE Linux Enterprise Server automatically logs almost everything that happens on the system i…

2 System Monitoring Utilities

There are number of programs, tools, and utilities which you can use to examine the status of your system. This chapter introduces some and describes their most important and frequently used parameters.

Note
Note: Gathering and Analyzing System Information with supportconfig

Apart from the utilities presented in the following, SUSE Linux Enterprise Server also contains supportconfig, a tool to create reports about the system such as: current kernel version, hardware, installed packages, partition setup and much more. These reports are used to provide the SUSE support with needed information in case a support ticket is created. However, they can also be analyzed for known issues to help resolve problems faster. For this purpose, SUSE Linux Enterprise Server provides both an appliance and a command line tool for Supportconfig Analysis (SCA). See Book “Administration Guide”, Chapter 41 “Gathering System Information for Support” for details.

For each of the described commands, examples of the relevant outputs are presented. In the examples, the first line is the command itself (after the tux > or root #). Omissions are indicated with square brackets ([...]) and long lines are wrapped where necessary. Line breaks for long lines are indicated by a backslash (\).

tux > command -x -y
output line 1
output line 2
output line 3 is annoyingly long, so long that \
    we need to break it
output line 4
[...]
output line 98
output line 99

The descriptions have been kept short so that we can include as many utilities as possible. Further information for all the commands can be found in the manual pages. Most of the commands also understand the parameter --help, which produces a brief list of possible parameters.

2.1 Multi-Purpose Tools

While most Linux system monitoring tools monitor only a single aspect of the system, there are a few tools with a broader scope. To get an overview and find out which part of the system to examine further, use these tools first.

2.1.1 vmstat

vmstat collects information about processes, memory, I/O, interrupts and CPU. If called without a sampling rate, it displays average values since the last reboot. When called with a sampling rate, it displays actual samples:

Example 2.1: vmstat Output on a Lightly Used Machine
tux > vmstat 2
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 1  0  44264  81520    424 935736    0    0    12    25   27   34  1  0 98   0  0
 0  0  44264  81552    424 935736    0    0     0     0   38   25  0  0 100  0  0
 0  0  44264  81520    424 935732    0    0     0     0   23   15  0  0 100  0  0
 0  0  44264  81520    424 935732    0    0     0     0   36   24  0  0 100  0  0
 0  0  44264  81552    424 935732    0    0     0     0   51   38  0  0 100  0  0
Example 2.2: vmstat Output on a Heavily Used Machine (CPU bound)
tux > vmstat 2
procs -----------memory----------- ---swap-- -----io---- -system-- -----cpu------
 r  b   swpd   free   buff   cache   si   so    bi    bo   in   cs us sy id wa st
32  1  26236 459640 110240 6312648    0    0  9944     2 4552 6597 95  5  0  0  0
23  1  26236 396728 110336 6136224    0    0  9588     0 4468 6273 94  6  0  0  0
35  0  26236 554920 110508 6166508    0    0  7684 27992 4474 4700 95  5  0  0  0
28  0  26236 518184 110516 6039996    0    0 10830     4 4446 4670 94  6  0  0  0
21  5  26236 716468 110684 6074872    0    0  8734 20534 4512 4061 96  4  0  0  0
Tip
Tip: First Line of Output

The first line of the vmstat output always displays average values since the last reboot.

The columns show the following:

r

Shows the number of processes in a runnable state. These processes are either executing or waiting for a free CPU slot. If the number of processes in this column is constantly higher than the number of CPUs available, this may be an indication of insufficient CPU power.

b

Shows the number of processes waiting for a resource other than a CPU. A high number in this column may indicate an I/O problem (network or disk).

swpd

The amount of swap space (KB) currently used.

free

The amount of unused memory (KB).

inact

Recently unused memory that can be reclaimed. This column is only visible when calling vmstat with the parameter -a (recommended).

active

Recently used memory that normally does not get reclaimed. This column is only visible when calling vmstat with the parameter -a (recommended).

buff

File buffer cache (KB) in RAM that contains file system metadata. This column is not visible when calling vmstat with the parameter -a.

cache

Page cache (KB) in RAM with the actual contents of files. This column is not visible when calling vmstat with the parameter -a.

si / so

Amount of data (KB) that is moved from swap to RAM (si) or from RAM to swap (so) per second. High so values over a long period of time may indicate that an application is leaking memory and the leaked memory is being swapped out. High si values over a long period of time could mean that an application that was inactive for a very long time is now active again. Combined high si and so values for prolonged periods of time are evidence of swap thrashing and may indicate that more RAM needs to be installed in the system because there is not enough memory to hold the working set size.

bi

Number of blocks per second received from a block device (for example, a disk read). Note that swapping also impacts the values shown here. The block size may vary between file systems but can be determined using the stat utility. If throughput data is required then iostat may be used.

bo

Number of blocks per second sent to a block device (for example, a disk write). Note that swapping also impacts the values shown here.

in

Interrupts per second. A high value may indicate a high I/O level (network and/or disk), but could also be triggered for other reasons such as inter-processor interrupts triggered by another activity. Make sure to also check /proc/interrupts to identify the source of interrupts.

cs

Number of context switches per second. This is the number of times that the kernel replaces executable code of one program in memory with that of another program.

us

Percentage of CPU usage executing application code.

sy

Percentage of CPU usage executing kernel code.

id

Percentage of CPU time spent idling. If this value is zero over a longer time, your CPU(s) are working to full capacity. This is not necessarily a bad sign—rather refer to the values in columns r and b to determine if your machine is equipped with sufficient CPU power.

wa

If "wa" time is non-zero, it indicates throughput lost because of waiting for I/O. This may be inevitable, for example, if a file is being read for the first time, background writeback cannot keep up, and so on. It can also be an indicator for a hardware bottleneck (network or hard disk). Lastly, it can indicate a potential for tuning the virtual memory manager (refer to Chapter 14, Tuning the Memory Management Subsystem).

st

Percentage of CPU time stolen from a virtual machine.

See vmstat --help for more options.

2.1.2 dstat

dstat is a replacement for tools such as vmstat, iostat, netstat, or ifstat. dstat displays information about the system resources in real time. For example, you can compare disk usage in combination with interrupts from the IDE controller, or compare network bandwidth with the disk throughput (in the same interval).

By default, its output is presented in readable tables. Alternatively, CSV output can be produced which is suitable as a spreadsheet import format.

It is written in Python and can be enhanced with plug-ins.

This is the general syntax:

dstat [-afv] [OPTIONS..] [DELAY [COUNT]]

All options and parameters are optional. Without any parameter, dstat displays statistics about CPU (-c, --cpu), disk (-d, --disk), network (-n, --net), paging (-g, --page), and the interrupts and context switches of the system (-y, --sys); it refreshes the output every second ad infinitum:

root # dstat
You did not select any stats, using -cdngy by default.
----total-cpu-usage---- -dsk/total- -net/total- ---paging-- ---system--
usr sys idl wai hiq siq| read  writ| recv  send|  in   out | int   csw
  0   0 100   0   0   0|  15k   44k|   0     0 |   0    82B| 148   194
  0   0 100   0   0   0|   0     0 |5430B  170B|   0     0 | 163   187
  0   0 100   0   0   0|   0     0 |6363B  842B|   0     0 | 196   185
-a, --all

equal to -cdngy (default)

-f, --full

expand -C, -D, -I, -N and -S discovery lists

-v, --vmstat

equal to -pmgdsc, -D total

DELAY

delay in seconds between each update

COUNT

the number of updates to display before exiting

The default delay is 1 and the count is unspecified (unlimited).

For more information, see the man page of dstat and its Web page at http://dag.wieers.com/home-made/dstat/.

2.1.3 System Activity Information: sar

sar can generate extensive reports on almost all important system activities, among them CPU, memory, IRQ usage, IO, or networking. It can also generate reports on the fly. sar gathers all their data from the /proc file system.

Note
Note: sysstat Package

sar is a part of the sysstat package either with YaST, or with zypper in sysstat.

2.1.3.1 Generating reports with sar

To generate reports on the fly, call sar with an interval (seconds) and a count. To generate reports from files specify a file name with the option -f instead of interval and count. If file name, interval and count are not specified, sar attempts to generate a report from /var/log/sa/saDD, where DD stands for the current day. This is the default location to where sadc (the system activity data collector) writes its data. Query multiple files with multiple -f options.

sar 2 10                         # on-the-fly report, 10 times every 2 seconds
sar -f ~/reports/sar_2014_07_17  # queries file sar_2014_07_17
sar                              # queries file from today in /var/log/sa/
cd /var/log/sa && \
sar -f sa01 -f sa02              # queries files /var/log/sa/0[12]

Find examples for useful sar calls and their interpretation below. For detailed information on the meaning of each column, refer to the man (1) of sar.

Note
Note: sysstat Reporting When the Service Stops

When the sysstat service is stopped (for example, during reboot or shutdown), the tool still collects last-minute statistics by automatically running the /usr/lib64/sa/sa1 -S ALL 1 1 command. The collected binary data is stored in the system activity data file.

2.1.3.1.1 CPU Usage Report: sar

When called with no options, sar shows a basic report about CPU usage. On multi-processor machines, results for all CPUs are summarized. Use the option -P ALL to also see statistics for individual CPUs.

root # sar 10 5
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

17:51:29        CPU     %user     %nice   %system   %iowait    %steal     %idle
17:51:39        all     57,93      0,00      9,58      1,01      0,00     31,47
17:51:49        all     32,71      0,00      3,79      0,05      0,00     63,45
17:51:59        all     47,23      0,00      3,66      0,00      0,00     49,11
17:52:09        all     53,33      0,00      4,88      0,05      0,00     41,74
17:52:19        all     56,98      0,00      5,65      0,10      0,00     37,27
Average:        all     49,62      0,00      5,51      0,24      0,00     44,62

%iowait displays the percentage of time that the CPU was idle while waiting for an I/O request. If this value is significantly higher than zero over a longer time, there is a bottleneck in the I/O system (network or hard disk). If the %idle value is zero over a longer time, your CPU is working at capacity.

2.1.3.1.2 Memory Usage Report: sar -r

Generate an overall picture of the system memory (RAM) by using the option -r:

root # sar -r 10 5
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

17:55:27 kbmemfree kbmemused %memused kbbuffers kbcached kbcommit %commit kbactive kbinact kbdirty
17:55:37    104232   1834624    94.62        20   627340  2677656   66.24   802052  828024    1744
17:55:47     98584   1840272    94.92        20   624536  2693936   66.65   808872  826932    2012
17:55:57     87088   1851768    95.51        20   605288  2706392   66.95   827260  821304    1588
17:56:07     86268   1852588    95.55        20   599240  2739224   67.77   829764  820888    3036
17:56:17    104260   1834596    94.62        20   599864  2730688   67.56   811284  821584    3164
Average:     96086   1842770    95.04        20   611254  2709579   67.03   815846  823746    2309

The columns kbcommit and %commit show an approximation of the maximum amount of memory (RAM and swap) that the current workload could need. While kbcommit displays the absolute number in kilobytes, %commit displays a percentage.

2.1.3.1.3 Paging Statistics Report: sar -B

Use the option -B to display the kernel paging statistics.

root # sar -B 10 5
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

18:23:01 pgpgin/s pgpgout/s fault/s majflt/s pgfree/s pgscank/s pgscand/s pgsteal/s %vmeff
18:23:11   366.80     11.60  542.50     1.10  4354.80      0.00      0.00      0.00   0.00
18:23:21     0.00    333.30 1522.40     0.00 18132.40      0.00      0.00      0.00   0.00
18:23:31    47.20    127.40 1048.30     0.10 11887.30      0.00      0.00      0.00   0.00
18:23:41    46.40      2.50  336.10     0.10  7945.00      0.00      0.00      0.00   0.00
18:23:51     0.00    583.70 2037.20     0.00 17731.90      0.00      0.00      0.00   0.00
Average:    92.08    211.70 1097.30     0.26 12010.28      0.00      0.00      0.00   0.00

The majflt/s (major faults per second) column shows how many pages are loaded from disk into memory. The source of the faults may be file accesses or faults. At times, many major faults are normal. For example, during application start-up time. If major faults are experienced for the entire lifetime of the application it may be an indication that there is insufficient main memory, particularly if combined with large amounts of direct scanning (pgscand/s).

The %vmeff column shows the number of pages scanned (pgscand/s) in relation to the ones being reused from the main memory cache or the swap cache (pgsteal/s). It is a measurement of the efficiency of page reclaim. Healthy values are either near 100 (every inactive page swapped out is being reused) or 0 (no pages have been scanned). The value should not drop below 30.

2.1.3.1.4 Block Device Statistics Report: sar -d

Use the option -d to display the block device (hard disk, optical drive, USB storage device, etc.). Make sure to use the additional option -p (pretty-print) to make the DEV column readable.

root # sar -d -p 10 5
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

18:46:09 DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
18:46:19 sda  1.70    33.60      0.00     19.76      0.00      0.47      0.47      0.08
18:46:19 sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

18:46:19 DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
18:46:29 sda  8.60   114.40    518.10     73.55      0.06      7.12      0.93      0.80
18:46:29 sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

18:46:29 DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
18:46:39 sda 40.50  3800.80    454.90    105.08      0.36      8.86      0.69      2.80
18:46:39 sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

18:46:39 DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
18:46:49 sda  1.40     0.00    204.90    146.36      0.00      0.29      0.29      0.04
18:46:49 sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

18:46:49 DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
18:46:59 sda  3.30     0.00    503.80    152.67      0.03      8.12      1.70      0.56
18:46:59 sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

Average: DEV   tps rd_sec/s  wr_sec/s  avgrq-sz  avgqu-sz     await     svctm     %util
Average: sda 11.10   789.76    336.34    101.45      0.09      8.07      0.77      0.86
Average: sr0  0.00     0.00      0.00      0.00      0.00      0.00      0.00      0.00

Compare the Average values for tps, rd_sec/s, and wr_sec/s of all disks. Constantly high values in the svctm and %util columns could be an indication that I/O subsystem is a bottleneck.

If the machine uses multiple disks, then it is best if I/O is interleaved evenly between disks of equal speed and capacity. It will be necessary to take into account whether the storage has multiple tiers. Furthermore, if there are multiple paths to storage then consider what the link saturation will be when balancing how storage is used.

2.1.3.1.5 Network Statistics Reports: sar -n KEYWORD

The option -n lets you generate multiple network related reports. Specify one of the following keywords along with the -n:

  • DEV: Generates a statistic report for all network devices

  • EDEV: Generates an error statistics report for all network devices

  • NFS: Generates a statistic report for an NFS client

  • NFSD: Generates a statistic report for an NFS server

  • SOCK: Generates a statistic report on sockets

  • ALL: Generates all network statistic reports

2.1.3.2 Visualizing sar Data with sadf

Reports generated by sar can sometimes be difficult to read. To make the data generated by sar easier to analyze, you can visualize the data using the sadf command. The -g parameter instructs sadf to output the visualized data in the SVG format. The following example sadf saves the visualized CPU utilization data in the output.sg file:

tux > sadf -g > output.svg

2.2 System Information

2.2.1 Device Load Information: iostat

To monitor the system device load, use iostat. It generates reports that can be useful for better balancing the load between physical disks attached to your system.

To be able to use iostat, install the package sysstat.

The first iostat report shows statistics collected since the system was booted. Subsequent reports cover the time since the previous report.

tux > iostat
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (4 CPU)

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
          17.68    4.49    4.24    0.29    0.00   73.31

Device:            tps    kB_read/s    kB_wrtn/s    kB_read    kB_wrtn
sdb               2.02        36.74        45.73    3544894    4412392
sda               1.05         5.12        13.47     493753    1300276
sdc               0.02         0.14         0.00      13641         37

Invoking iostat in this way will help you find out whether throughput is different from your expectation, but not why. Such questions can be better answered by an extended report which can be generated by invoking iostat -x. Extended reports additionally include, for example, information on average queue sizes and average wait times. It may also be easier to evaluate the data if idle block devices are excluded using the -z switch. Find definitions for each of the displayed column titles in the man page of iostat (man 1 iostat).

You can also specify that a certain device should be monitored at specified intervals. For example, to generate five reports at three-second intervals for the device sda, use:

tux > iostat -p sda 3 5

To show statistics of network file systems (NFS), there are two similar utilities:

  • nfsiostat-sysstat is included with the package sysstat.

  • nfsiostat is included with the package nfs-client.

2.2.2 Processor Activity Monitoring: mpstat

The utility mpstat examines activities of each available processor. If your system has one processor only, the global average statistics will be reported.

The timing arguments work the same way as with the iostat command. Entering mpstat 2 5 prints five reports for all processors in two-second intervals.

root # mpstat 2 5
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

13:51:10  CPU   %usr  %nice  %sys  %iowait  %irq  %soft  %steal  %guest  %gnice   %idle
13:51:12  all   8,27   0,00  0,50     0,00  0,00   0,00    0,00    0,00    0,00   91,23
13:51:14  all  46,62   0,00  3,01     0,00  0,00   0,25    0,00    0,00    0,00   50,13
13:51:16  all  54,71   0,00  3,82     0,00  0,00   0,51    0,00    0,00    0,00   40,97
13:51:18  all  78,77   0,00  5,12     0,00  0,00   0,77    0,00    0,00    0,00   15,35
13:51:20  all  51,65   0,00  4,30     0,00  0,00   0,51    0,00    0,00    0,00   43,54
Average:  all  47,85   0,00  3,34     0,00  0,00   0,40    0,00    0,00    0,00   48,41

From the mpstat data, you can see:

  • The ratio between the %usr and %sys. For example, a ratio of 10:1 indicates the workload is mostly running application code and analysis should focus on the application. A ratio of 1:10 indicates the workload is mostly kernel-bound and tuning the kernel is worth considering. Alternatively, determine why the application is kernel-bound and see if that can be alleviated.

  • Whether there is a subset of CPUs that are nearly fully utilized even if the system is lightly loaded overall. Few hot CPUs can indicate that the workload is not parallelized and could benefit from executing on a machine with a smaller number of faster processors.

2.2.3 Processor Frequency Monitoring: turbostat

turbostat shows frequencies, load, temperature, and power of AMD64/Intel 64 processors. It can operate in two modes: If called with a command, the command process is forked and statistics are displayed upon command completion. When run without a command, it will display updated statistics every five seconds. Note that turbostat requires the kernel module msr to be loaded.

tux > sudo turbostat find /etc -type d -exec true {} \;
0.546880 sec
     CPU Avg_MHz   Busy% Bzy_MHz TSC_MHz
       -     416   28.43    1465    3215
       0     631   37.29    1691    3215
       1     416   27.14    1534    3215
       2     270   24.30    1113    3215
       3     406   26.57    1530    3214
       4     505   32.46    1556    3214
       5     270   22.79    1184    3214

The output depends on the CPU type and may vary. To display more details such as temperature and power, use the --debug option. For more command line options and an explanation of the field descriptions, refer to man 8 turbostat.

2.2.4 Task Monitoring: pidstat

If you need to see what load a particular task applies to your system, use pidstat command. It prints activity of every selected task or all tasks managed by Linux kernel if no task is specified. You can also set the number of reports to be displayed and the time interval between them.

For example, pidstat -C firefox 2 3 prints the load statistic for tasks whose command name includes the string firefox. There will be three reports printed at two second intervals.

root # pidstat -C firefox 2 3
Linux 4.12.14-94.37-default (jupiter)         10/12/16        _x86_64_        (2 CPU)

14:09:11      UID       PID    %usr %system  %guest    %CPU   CPU  Command
14:09:13     1000       387   22,77    0,99    0,00   23,76     1  firefox

14:09:13      UID       PID    %usr %system  %guest    %CPU   CPU  Command
14:09:15     1000       387   46,50    3,00    0,00   49,50     1  firefox

14:09:15      UID       PID    %usr %system  %guest    %CPU   CPU  Command
14:09:17     1000       387   60,50    7,00    0,00   67,50     1  firefox

Average:      UID       PID    %usr %system  %guest    %CPU   CPU  Command
Average:     1000       387   43,19    3,65    0,00   46,84     -  firefox

Similarly, pidstat -d can be used to estimate how much I/O tasks are doing, whether they are sleeping on that I/O and how many clock ticks the task was stalled.

2.2.5 Kernel Ring Buffer: dmesg

The Linux kernel keeps certain messages in a ring buffer. To view these messages, enter the command dmesg -T.

Older events are logged in the systemd journal. See Book “Administration Guide”, Chapter 16 “journalctl: Query the systemd Journal” for more information on the journal.

2.2.6 List of Open Files: lsof

To view a list of all the files open for the process with process ID PID, use -p. For example, to view all the files used by the current shell, enter:

root # lsof -p $$
COMMAND  PID USER   FD   TYPE DEVICE SIZE/OFF  NODE NAME
bash    8842 root  cwd    DIR   0,32      222  6772 /root
bash    8842 root  rtd    DIR   0,32      166   256 /
bash    8842 root  txt    REG   0,32   656584 31066 /bin/bash
bash    8842 root  mem    REG   0,32  1978832 22993 /lib64/libc-2.19.so
[...]
bash    8842 root    2u   CHR  136,2      0t0     5 /dev/pts/2
bash    8842 root  255u   CHR  136,2      0t0     5 /dev/pts/2

The special shell variable $$, whose value is the process ID of the shell, has been used.

When used with -i, lsof lists currently open Internet files as well:

root # lsof -i
COMMAND    PID USER   FD   TYPE DEVICE SIZE/OFF NODE NAME
wickedd-d  917 root    8u  IPv4  16627      0t0  UDP *:bootpc
wickedd-d  918 root    8u  IPv6  20752      0t0  UDP [fe80::5054:ff:fe72:5ead]:dhcpv6-client
sshd      3152 root    3u  IPv4  18618      0t0  TCP *:ssh (LISTEN)
sshd      3152 root    4u  IPv6  18620      0t0  TCP *:ssh (LISTEN)
master    4746 root   13u  IPv4  20588      0t0  TCP localhost:smtp (LISTEN)
master    4746 root   14u  IPv6  20589      0t0  TCP localhost:smtp (LISTEN)
sshd      8837 root    5u  IPv4 293709      0t0  TCP jupiter.suse.de:ssh->venus.suse.de:33619 (ESTABLISHED)
sshd      8837 root    9u  IPv6 294830      0t0  TCP localhost:x11 (LISTEN)
sshd      8837 root   10u  IPv4 294831      0t0  TCP localhost:x11 (LISTEN)

2.2.7 Kernel and udev Event Sequence Viewer: udevadm monitor

udevadm monitor listens to the kernel uevents and events sent out by a udev rule and prints the device path (DEVPATH) of the event to the console. This is a sequence of events while connecting a USB memory stick:

Note
Note: Monitoring udev Events

Only root user is allowed to monitor udev events by running the udevadm command.

UEVENT[1138806687] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2
UEVENT[1138806687] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2/4-2.2
UEVENT[1138806687] add@/class/scsi_host/host4
UEVENT[1138806687] add@/class/usb_device/usbdev4.10
UDEV  [1138806687] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2
UDEV  [1138806687] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2/4-2.2
UDEV  [1138806687] add@/class/scsi_host/host4
UDEV  [1138806687] add@/class/usb_device/usbdev4.10
UEVENT[1138806692] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2/4-2.2
UEVENT[1138806692] add@/block/sdb
UEVENT[1138806692] add@/class/scsi_generic/sg1
UEVENT[1138806692] add@/class/scsi_device/4:0:0:0
UDEV  [1138806693] add@/devices/pci0000:00/0000:00:1d.7/usb4/4-2/4-2.2/4-2.2
UDEV  [1138806693] add@/class/scsi_generic/sg1
UDEV  [1138806693] add@/class/scsi_device/4:0:0:0
UDEV  [1138806693] add@/block/sdb
UEVENT[1138806694] add@/block/sdb/sdb1
UDEV  [1138806694] add@/block/sdb/sdb1
UEVENT[1138806694] mount@/block/sdb/sdb1
UEVENT[1138806697] umount@/block/sdb/sdb1

2.3 Processes

2.3.1 Interprocess Communication: ipcs

The command ipcs produces a list of the IPC resources currently in use:

root # ipcs
------ Message Queues --------
key        msqid      owner      perms      used-bytes   messages

------ Shared Memory Segments --------
key        shmid      owner      perms      bytes      nattch     status
0x00000000 65536      tux        600        524288     2          dest
0x00000000 98305      tux        600        4194304    2          dest
0x00000000 884738     root       600        524288     2          dest
0x00000000 786435     tux        600        4194304    2          dest
0x00000000 12058628   tux        600        524288     2          dest
0x00000000 917509     root       600        524288     2          dest
0x00000000 12353542   tux        600        196608     2          dest
0x00000000 12451847   tux        600        524288     2          dest
0x00000000 11567114   root       600        262144     1          dest
0x00000000 10911763   tux        600        2097152    2          dest
0x00000000 11665429   root       600        2336768    2          dest
0x00000000 11698198   root       600        196608     2          dest
0x00000000 11730967   root       600        524288     2          dest

------ Semaphore Arrays --------
key        semid      owner      perms      nsems
0xa12e0919 32768      tux        666        2

2.3.2 Process List: ps

The command ps produces a list of processes. Most parameters must be written without a minus sign. Refer to ps --help for a brief help or to the man page for extensive help.

To list all processes with user and command line information, use ps axu:

tux > ps axu
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root         1  0.0  0.3  34376  4608 ?        Ss   Jul24   0:02 /usr/lib/systemd/systemd
root         2  0.0  0.0      0     0 ?        S    Jul24   0:00 [kthreadd]
root         3  0.0  0.0      0     0 ?        S    Jul24   0:00 [ksoftirqd/0]
root         5  0.0  0.0      0     0 ?        S<   Jul24   0:00 [kworker/0:0H]
root         6  0.0  0.0      0     0 ?        S    Jul24   0:00 [kworker/u2:0]
root         7  0.0  0.0      0     0 ?        S    Jul24   0:00 [migration/0]
[...]
tux      12583  0.0  0.1 185980  2720 ?        Sl   10:12   0:00 /usr/lib/gvfs/gvfs-mtp-volume-monitor
tux      12587  0.0  0.1 198132  3044 ?        Sl   10:12   0:00 /usr/lib/gvfs/gvfs-gphoto2-volume-monitor
tux      12591  0.0  0.1 181940  2700 ?        Sl   10:12   0:00 /usr/lib/gvfs/gvfs-goa-volume-monitor
tux      12594  8.1 10.6 1418216 163564 ?      Sl   10:12   0:03 /usr/bin/gnome-shell
tux      12600  0.0  0.3 393448  5972 ?        Sl   10:12   0:00 /usr/lib/gnome-settings-daemon-3.0/gsd-printer
tux      12625  0.0  0.6 227776 10112 ?        Sl   10:12   0:00 /usr/lib/gnome-control-center-search-provider
tux      12626  0.5  1.5 890972 23540 ?        Sl   10:12   0:00 /usr/bin/nautilus --no-default-window
[...]

To check how many sshd processes are running, use the option -p together with the command pidof, which lists the process IDs of the given processes.

tux > ps -p $(pidof sshd)
  PID TTY      STAT   TIME COMMAND
 1545 ?        Ss     0:00 /usr/sbin/sshd -D
 4608 ?        Ss     0:00 sshd: root@pts/0

The process list can be formatted according to your needs. The option L returns a list of all keywords. Enter the following command to issue a list of all processes sorted by memory usage:

tux > ps ax --format pid,rss,cmd --sort rss
  PID   RSS CMD
  PID   RSS CMD
    2     0 [kthreadd]
    3     0 [ksoftirqd/0]
    4     0 [kworker/0:0]
    5     0 [kworker/0:0H]
    6     0 [kworker/u2:0]
    7     0 [migration/0]
    8     0 [rcu_bh]
[...]
12518 22996 /usr/lib/gnome-settings-daemon-3.0/gnome-settings-daemon
12626 23540 /usr/bin/nautilus --no-default-window
12305 32188 /usr/bin/Xorg :0 -background none -verbose
12594 164900 /usr/bin/gnome-shell
Useful ps Calls
ps aux--sort COLUMN

Sort the output by COLUMN. Replace COLUMN with

pmem for physical memory ratio
pcpu for CPU ratio
rss for resident set size (non-swapped physical memory)
ps axo pid,%cpu,rss,vsz,args,wchan

Shows every process, their PID, CPU usage ratio, memory size (resident and virtual), name, and their syscall.

ps axfo pid,args

Show a process tree.

2.3.3 Process Tree: pstree

The command pstree produces a list of processes in the form of a tree:

tux > pstree
systemd---accounts-daemon---{gdbus}
        |                 |-{gmain}
        |-at-spi-bus-laun---dbus-daemon
        |                 |-{dconf worker}
        |                 |-{gdbus}
        |                 |-{gmain}
        |-at-spi2-registr---{gdbus}
        |-cron
        |-2*[dbus-daemon]
        |-dbus-launch
        |-dconf-service---{gdbus}
        |               |-{gmain}
        |-gconfd-2
        |-gdm---gdm-simple-slav---Xorg
        |     |                 |-gdm-session-wor---gnome-session---gnome-setti+
        |     |                 |                 |               |-gnome-shell+++
        |     |                 |                 |               |-{dconf work+
        |     |                 |                 |               |-{gdbus}
        |     |                 |                 |               |-{gmain}
        |     |                 |                 |-{gdbus}
        |     |                 |                 |-{gmain}
        |     |                 |-{gdbus}
        |     |                 |-{gmain}
        |     |-{gdbus}
        |     |-{gmain}
[...]

The parameter -p adds the process ID to a given name. To have the command lines displayed as well, use the -a parameter:

2.3.4 Table of Processes: top

The command top (an abbreviation of table of processes) displays a list of processes that is refreshed every two seconds. To terminate the program, press q. The parameter -n 1 terminates the program after a single display of the process list. The following is an example output of the command top -n 1:

tux > top -n 1
Tasks: 128 total,   1 running, 127 sleeping,   0 stopped,   0 zombie
%Cpu(s):  2.4 us,  1.2 sy,  0.0 ni, 96.3 id,  0.1 wa,  0.0 hi,  0.0 si,  0.0 st
KiB Mem:   1535508 total,   699948 used,   835560 free,      880 buffers
KiB Swap:  1541116 total,        0 used,  1541116 free.   377000 cached Mem

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
    1 root      20   0  116292   4660   2028 S 0.000 0.303   0:04.45 systemd
    2 root      20   0       0      0      0 S 0.000 0.000   0:00.00 kthreadd
    3 root      20   0       0      0      0 S 0.000 0.000   0:00.07 ksoftirqd+
    5 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 kworker/0+
    6 root      20   0       0      0      0 S 0.000 0.000   0:00.00 kworker/u+
    7 root      rt   0       0      0      0 S 0.000 0.000   0:00.00 migration+
    8 root      20   0       0      0      0 S 0.000 0.000   0:00.00 rcu_bh
    9 root      20   0       0      0      0 S 0.000 0.000   0:00.24 rcu_sched
   10 root      rt   0       0      0      0 S 0.000 0.000   0:00.01 watchdog/0
   11 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 khelper
   12 root      20   0       0      0      0 S 0.000 0.000   0:00.00 kdevtmpfs
   13 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 netns
   14 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 writeback
   15 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 kintegrit+
   16 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 bioset
   17 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 crypto
   18 root       0 -20       0      0      0 S 0.000 0.000   0:00.00 kblockd

By default the output is sorted by CPU usage (column %CPU, shortcut ShiftP). Use the following key combinations to change the sort field:

ShiftM: Resident Memory (RES)
ShiftN: Process ID (PID)
ShiftT: Time (TIME+)

To use any other field for sorting, press F and select a field from the list. To toggle the sort order, Use ShiftR.

The parameter -U UID monitors only the processes associated with a particular user. Replace UID with the user ID of the user. Use top -U $(id -u) to show processes of the current user

2.3.5 IBM IBM Z Hypervisor Monitor: hyptop

hyptop provides a dynamic real-time view of a IBM IBM Z hypervisor environment, using the kernel infrastructure via debugfs. It works with either the z/VM or the LPAR hypervisor. Depending on the available data it, for example, shows CPU and memory consumption of active LPARs or z/VM guests. It provides a curses based user interface similar to the top command. hyptop provides two windows:

  • sys_list: Lists systems that the currently hypervisor is running

  • sys: Shows one system in more detail

You can run hyptop in interactive mode (default) or in batch mode with the -b option. Help in the interactive mode is available by pressing ? after hyptop is started.

Output for the sys_list window under LPAR:

12:30:48 | CPU-T: IFL(18) CP(3) UN(3)     ?=help
system  #cpu    cpu   mgm    Cpu+  Mgm+   online
(str)    (#)    (%)   (%)    (hm)  (hm)    (dhm)
H05LP30   10 461.14 10.18 1547:41  8:15 11:05:59
H05LP33    4 133.73  7.57  220:53  6:12 11:05:54
H05LP50    4  99.26  0.01  146:24  0:12 10:04:24
H05LP02    1  99.09  0.00  269:57  0:00 11:05:58
TRX2CFA    1   2.14  0.03    3:24  0:04 11:06:01
H05LP13    6   1.36  0.34    4:23  0:54 11:05:56
TRX1      19   1.22  0.14   13:57  0:22 11:06:01
TRX2      20   1.16  0.11   26:05  0:25 11:06:00
H05LP55    2   0.00  0.00    0:22  0:00 11:05:52
H05LP56    3   0.00  0.00    0:00  0:00 11:05:52
         413 823.39 23.86 3159:57 38:08 11:06:01

Output for the "sys_list" window under z/VM:

12:32:21 | CPU-T: UN(16)                          ?=help
system   #cpu    cpu    Cpu+   online memuse memmax wcur
(str)     (#)    (%)    (hm)    (dhm)  (GiB)  (GiB)  (#)
T6360004    6 100.31  959:47 53:05:20   1.56   2.00  100
T6360005    2   0.44    1:11  3:02:26   0.42   0.50  100
T6360014    2   0.27    0:45 10:18:41   0.54   0.75  100
DTCVSW1     1   0.00    0:00 53:16:42   0.01   0.03  100
T6360002    6   0.00  166:26 40:19:18   1.87   2.00  100
OPERATOR    1   0.00    0:00 53:16:42   0.00   0.03  100
T6360008    2   0.00    0:37 30:22:55   0.32   0.75  100
T6360003    6   0.00 3700:57 53:03:09   4.00   4.00  100
NSLCF1      1   0.00    0:02 53:16:41   0.03   0.25  500
EREP        1   0.00    0:00 53:16:42   0.00   0.03  100
PERFSVM     1   0.00    0:53  2:21:12   0.04   0.06    0
TCPIP       1   0.00    0:01 53:16:42   0.01   0.12 3000
DATAMOVE    1   0.00    0:05 53:16:42   0.00   0.03  100
DIRMAINT    1   0.00    0:04 53:16:42   0.01   0.03  100
DTCVSW2     1   0.00    0:00 53:16:42   0.01   0.03  100
RACFVM      1   0.00    0:00 53:16:42   0.01   0.02  100
           75 101.57 5239:47 53:16:42  15.46  22.50 3000

Output for the sys window under LPAR:

14:08:41 | H05LP30 | CPU-T: IFL(18) CP(3) UN(3)                  ? = help
cpuid   type    cpu   mgm visual.
(#)    (str)    (%)   (%) (vis)
0        IFL  96.91  1.96 |############################################ |
1        IFL  81.82  1.46 |#####################################        |
2        IFL  88.00  2.43 |########################################     |
3        IFL  92.27  1.29 |##########################################   |
4        IFL  83.32  1.05 |#####################################        |
5        IFL  92.46  2.59 |##########################################   |
6        IFL   0.00  0.00 |                                             |
7        IFL   0.00  0.00 |                                             |
8        IFL   0.00  0.00 |                                             |
9        IFL   0.00  0.00 |                                             |
             534.79 10.78

Output for the sys window under z/VM:

15:46:57 | T6360003 | CPU-T: UN(16)                  ? = help
cpuid     cpu visual
(#)       (%) (vis)
0      548.72 |#########################################    |
        548.72

2.3.6 A top-like I/O Monitor: iotop

The iotop utility displays a table of I/O usage by processes or threads.

Note
Note: Installing iotop

iotop is not installed by default. You need to install it manually with zypper in iotop as root.

iotop displays columns for the I/O bandwidth read and written by each process during the sampling period. It also displays the percentage of time the process spent while swapping in and while waiting on I/O. For each process, its I/O priority (class/level) is shown. In addition, the total I/O bandwidth read and written during the sampling period is displayed at the top of the interface.

  • The and keys change the sorting.

  • R reverses the sort order.

  • O toggles between showing all processes and threads (default view) and showing only those doing I/O. (This function is similar to adding --only on command line.)

  • P toggles between showing threads (default view) and processes. (This function is similar to --only.)

  • A toggles between showing the current I/O bandwidth (default view) and accumulated I/O operations since iotop was started. (This function is similar to --accumulated.)

  • I lets you change the priority of a thread or a process's threads.

  • Q quits iotop.

  • Pressing any other key will force a refresh.

Following is an example output of the command iotop --only, while find and emacs are running:

root # iotop --only
Total DISK READ: 50.61 K/s | Total DISK WRITE: 11.68 K/s
  TID  PRIO  USER     DISK READ  DISK WRITE  SWAPIN     IO>    COMMAND
 3416 be/4 tux         50.61 K/s    0.00 B/s  0.00 %  4.05 % find /
  275 be/3 root        0.00 B/s    3.89 K/s  0.00 %  2.34 % [jbd2/sda2-8]
 5055 be/4 tux          0.00 B/s    3.89 K/s  0.00 %  0.04 % emacs

iotop can be also used in a batch mode (-b) and its output stored in a file for later analysis. For a complete set of options, see the manual page (man 8 iotop).

2.3.7 Modify a process's niceness: nice and renice

The kernel determines which processes require more CPU time than others by the process's nice level, also called niceness. The higher the nice level of a process is, the less CPU time it will take from other processes. Nice levels range from -20 (the least nice level) to 19. Negative values can only be set by root.

Adjusting the niceness level is useful when running a non time-critical process that lasts long and uses large amounts of CPU time. For example, compiling a kernel on a system that also performs other tasks. Making such a process nicer, ensures that the other tasks, for example a Web server, will have a higher priority.

Calling nice without any parameters prints the current niceness:

tux > nice
0

Running nice COMMAND increments the current nice level for the given command by 10. Using nice -n LEVEL COMMAND lets you specify a new niceness relative to the current one.

To change the niceness of a running process, use renice PRIORITY -p PROCESS_ID, for example:

tux > renice +5 3266

To renice all processes owned by a specific user, use the option -u USER. Process groups are reniced by the option -g PROCESS_GROUP_ID.

2.4 Memory

2.4.1 Memory Usage: free

The utility free examines RAM and swap usage. Details of both free and used memory and swap areas are shown:

tux > free
             total       used       free     shared    buffers     cached
Mem:      32900500   32703448     197052          0     255668    5787364
-/+ buffers/cache:   26660416    6240084
Swap:      2046972     304680    1742292

The options -b, -k, -m, -g show the output in bytes, KB, MB, or GB, respectively. The parameter -s delay ensures that the display is refreshed every DELAY seconds. For example, free -s 1.5 produces an update every 1.5 seconds.

2.4.2 Detailed Memory Usage: /proc/meminfo

Use /proc/meminfo to get more detailed information on memory usage than with free. Actually free uses some data from this file. See an example output from a 64-bit system below. Note that it slightly differs on 32-bit systems because of different memory management:

MemTotal:        1942636 kB
MemFree:         1294352 kB
MemAvailable:    1458744 kB
Buffers:             876 kB
Cached:           278476 kB
SwapCached:            0 kB
Active:           368328 kB
Inactive:         199368 kB
Active(anon):     288968 kB
Inactive(anon):    10568 kB
Active(file):      79360 kB
Inactive(file):   188800 kB
Unevictable:          80 kB
Mlocked:              80 kB
SwapTotal:       2103292 kB
SwapFree:        2103292 kB
Dirty:                44 kB
Writeback:             0 kB
AnonPages:        288592 kB
Mapped:            70444 kB
Shmem:             11192 kB
Slab:              40916 kB
SReclaimable:      17712 kB
SUnreclaim:        23204 kB
KernelStack:        2000 kB
PageTables:        10996 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:     3074608 kB
Committed_AS:    1407208 kB
VmallocTotal:   34359738367 kB
VmallocUsed:      145996 kB
VmallocChunk:   34359588844 kB
HardwareCorrupted:     0 kB
AnonHugePages:     86016 kB
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
DirectMap4k:       79744 kB
DirectMap2M:     2017280 kB

These entries stand for the following:

MemTotal

Total amount of RAM.

MemFree

Amount of unused RAM.

MemAvailable

Estimate of how much memory is available for starting new applications without swapping.

Buffers

File buffer cache in RAM containing file system metadata.

Cached

Page cache in RAM. This excludes buffer cache and swap cache, but includes Shmem memory.

SwapCached

Page cache for swapped-out memory.

Active, Active(anon), Active(file)

Recently used memory that will not be reclaimed unless necessary or on explicit request. Active is the sum of Active(anon) and Active(file):

  • Active(anon) tracks swap-backed memory. This includes private and shared anonymous mappings and private file pages after copy-on-write.

  • Active(file) tracks other file system backed memory.

Inactive, Inactive(anon), Inactive(file)

Less recently used memory that will usually be reclaimed first. Inactive is the sum of Inactive(anon) and Inactive(file):

  • Inactive(anon) tracks swap backed memory. This includes private and shared anonymous mappings and private file pages after copy-on-write.

  • Inactive(file) tracks other file system backed memory.

Unevictable

Amount of memory that cannot be reclaimed (for example, because it is Mlocked or used as a RAM disk).

Mlocked

Amount of memory that is backed by the mlock system call. mlock allows processes to define which part of physical RAM their virtual memory should be mapped to. However, mlock does not guarantee this placement.

SwapTotal

Amount of swap space.

SwapFree

Amount of unused swap space.

Dirty

Amount of memory waiting to be written to disk, because it contains changes compared to the backing storage. Dirty data can be explicitly synchronized either by the application or by the kernel after a short delay. A large amount of dirty data may take considerable time to write to disk resulting in stalls. The total amount of dirty data that can exist at any time can be controlled with the sysctl parameters vm.dirty_ratio or vm.dirty_bytes (refer to Section 14.1.5, “Writeback” for more details).

Writeback

Amount of memory that is currently being written to disk.

Mapped

Memory claimed with the mmap system call.

Shmem

Memory shared between groups of processes, such as IPC data, tmpfs data, and shared anonymous memory.

Slab

Memory allocation for internal data structures of the kernel.

SReclaimable

Slab section that can be reclaimed, such as caches (inode, dentry, etc.).

SUnreclaim

Slab section that cannot be reclaimed.

KernelStack

Amount of kernel space memory used by applications (through system calls).

PageTables

Amount of memory dedicated to page tables of all processes.

NFS_Unstable

NFS pages that have already been sent to the server, but are not yet committed there.

Bounce

Memory used for bounce buffers of block devices.

WritebackTmp

Memory used by FUSE for temporary writeback buffers.

CommitLimit

Amount of memory available to the system based on the overcommit ratio setting. This is only enforced if strict overcommit accounting is enabled.

Committed_AS

An approximation of the total amount of memory (RAM and swap) that the current workload would need in the worst case.

VmallocTotal

Amount of allocated kernel virtual address space.

VmallocUsed

Amount of used kernel virtual address space.

VmallocChunk

The largest contiguous block of available kernel virtual address space.

HardwareCorrupted

Amount of failed memory (can only be detected when using ECC RAM).

AnonHugePages

Anonymous hugepages that are mapped into user space page tables. These are allocated transparently for processes without being specifically requested, therefore they are also known as transparent hugepages (THP).

HugePages_Total

Number of preallocated hugepages for use by SHM_HUGETLB and MAP_HUGETLB or through the hugetlbfs file system, as defined in /proc/sys/vm/nr_hugepages.

HugePages_Free

Number of hugepages available.

HugePages_Rsvd

Number of hugepages that are committed.

HugePages_Surp

Number of hugepages available beyond HugePages_Total (surplus), as defined in /proc/sys/vm/nr_overcommit_hugepages.

Hugepagesize

Size of a hugepage—on AMD64/Intel 64 the default is 2048 KB.

DirectMap4k etc.

Amount of kernel memory that is mapped to pages with a given size (in the example: 4 kB).

2.4.3 Process Memory Usage: smaps

Exactly determining how much memory a certain process is consuming is not possible with standard tools like top or ps. Use the smaps subsystem, introduced in kernel 2.6.14, if you need exact data. It can be found at /proc/PID/smaps and shows you the number of clean and dirty memory pages the process with the ID PID is using at that time. It differentiates between shared and private memory, so you can see how much memory the process is using without including memory shared with other processes. For more information see /usr/src/linux/Documentation/filesystems/proc.txt (requires the package kernel-source to be installed).

smaps is expensive to read. Therefore it is not recommended to monitor it regularly, but only when closely monitoring a certain process.

2.5 Networking

Tip
Tip: Traffic Shaping

In case the network bandwidth is lower than expected, you should first check if any traffic shaping rules are active for your network segment.

2.5.1 Basic Network Diagnostics: ip

ip is a powerful tool to set up and control network interfaces. You can also use it to quickly view basic statistics about network interfaces of the system. For example, whether the interface is up or how many errors, dropped packets, or packet collisions there are.

If you run ip with no additional parameter, it displays a help output. To list all network interfaces, enter ip addr show (or abbreviated as ip a). ip addr show up lists only running network interfaces. ip -s link show DEVICE lists statistics for the specified interface only:

root # ip -s link show br0
6: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT
    link/ether 00:19:d1:72:d4:30 brd ff:ff:ff:ff:ff:ff
    RX: bytes  packets  errors  dropped overrun mcast
    6346104756 9265517  0       10860   0       0
    TX: bytes  packets  errors  dropped carrier collsns
    3996204683 3655523  0       0       0       0

ip can also show interfaces (link), routing tables (route), and much more—refer to man 8 ip for details.

root # ip route
default via 192.168.2.1 dev eth1
192.168.2.0/24 dev eth0  proto kernel  scope link  src 192.168.2.100
192.168.2.0/24 dev eth1  proto kernel  scope link  src 192.168.2.101
192.168.2.0/24 dev eth2  proto kernel  scope link  src 192.168.2.102
root # ip link
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN mode DEFAULT group default
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000
    link/ether 52:54:00:44:30:51 brd ff:ff:ff:ff:ff:ff
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000
    link/ether 52:54:00:a3:c1:fb brd ff:ff:ff:ff:ff:ff
4: eth2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000
    link/ether 52:54:00:32:a4:09 brd ff:ff:ff:ff:ff:ff

2.5.2 Show the Network Usage of Processes: nethogs

In some cases, for example if the network traffic suddenly becomes very high, it is desirable to quickly find out which application(s) is/are causing the traffic. nethogs, a tool with a design similar to top, shows incoming and outgoing traffic for all relevant processes:

PID   USER  PROGRAM                                DEV   SENT   RECEIVED
27145 root   zypper                                eth0  5.719  391.749 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30015        0.102    2.326 KB/sec
26635 tux    /usr/lib64/firefox/firefox            eth0  0.026    0.026 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30045        0.000    0.021 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30045        0.000    0.018 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30015        0.000    0.018 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30045        0.000    0.017 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30045        0.000    0.017 KB/sec
?     root   ..0:113:80c0:8080:10:160:0:100:30045        0.069    0.000 KB/sec
?     root   unknown TCP                                 0.000    0.000 KB/sec

TOTAL                                                  5.916  394.192 KB/sec

Like in top, nethogs features interactive commands:

M: cycle between display modes (kb/s, kb, b, mb)
R: sort by RECEIVED
S: sort by SENT
Q: quit

2.5.3 Ethernet Cards in Detail: ethtool

ethtool can display and change detailed aspects of your Ethernet network device. By default it prints the current setting of the specified device.

root # ethtool eth0
Settings for eth0:
 Supported ports: [ TP ]
 Supported link modes:   10baseT/Half 10baseT/Full
                         100baseT/Half 100baseT/Full
                         1000baseT/Full
 Supports auto-negotiation: Yes
 Advertised link modes:  10baseT/Half 10baseT/Full
                         100baseT/Half 100baseT/Full
                         1000baseT/Full
 Advertised pause frame use: No
[...]
 Link detected: yes

The following table shows ethtool options that you can use to query the device for specific information:

Table 2.1: List of Query Options of ethtool

ethtool option

it queries the device for

-a

pause parameter information

-c

interrupt coalescing information

-g

Rx/Tx (receive/transmit) ring parameter information

-i

associated driver information

-k

offload information

-S

NIC and driver-specific statistics

2.5.4 Show the Network Status: ss

ss is a tool to dump socket statistics and replaces the netstat command. To list all connections use ss without parameters:

root # ss
Netid  State      Recv-Q Send-Q   Local Address:Port       Peer Address:Port
u_str  ESTAB      0      0                    * 14082                 * 14083
u_str  ESTAB      0      0                    * 18582                 * 18583
u_str  ESTAB      0      0                    * 19449                 * 19450
u_str  ESTAB      0      0      @/tmp/dbus-gmUUwXABPV 18784           * 18783
u_str  ESTAB      0      0      /var/run/dbus/system_bus_socket 19383 * 19382
u_str  ESTAB      0      0      @/tmp/dbus-gmUUwXABPV 18617           * 18616
u_str  ESTAB      0      0      @/tmp/dbus-58TPPDv8qv 19352           * 19351
u_str  ESTAB      0      0                    * 17658                 * 17657
u_str  ESTAB      0      0                    * 17693                 * 17694
[..]

To show all network ports currently open, use the following command:

root # ss -l
Netid  State      Recv-Q Send-Q      Local Address:Port  Peer Address:Port
nl     UNCONN     0      0                 rtnl:4195117                  *
nl     UNCONN     0      0       rtnl:wickedd-auto4/811                  *
nl     UNCONN     0      0       rtnl:wickedd-dhcp4/813                  *
nl     UNCONN     0      0                 rtnl:4195121                  *
nl     UNCONN     0      0                 rtnl:4195115                  *
nl     UNCONN     0      0       rtnl:wickedd-dhcp6/814                  *
nl     UNCONN     0      0                  rtnl:kernel                  *
nl     UNCONN     0      0             rtnl:wickedd/817                  *
nl     UNCONN     0      0                 rtnl:4195118                  *
nl     UNCONN     0      0                rtnl:nscd/706                  *
nl     UNCONN     4352   0              tcpdiag:ss/2381                  *
[...]

When displaying network connections, you can specify the socket type to display: TCP (-t) or UDP (-u) for example. The -p option shows the PID and name of the program to which each socket belongs.

The following example lists all TCP connections and the programs using these connections. The -a option make sure all established connections (listening and non-listening) are shown. The -p option shows the PID and name of the program to which each socket belongs.

root # ss -t -a -p
State    Recv-Q Send-Q  Local Address:Port   Peer Address:Port
LISTEN   0      128                  *:ssh                 *:*  users:(("sshd",1551,3))
LISTEN   0      100         127.0.0.1:smtp                 *:*  users:(("master",1704,13))
ESTAB    0      132      10.120.65.198:ssh  10.120.4.150:55715  users:(("sshd",2103,5))
LISTEN   0      128                 :::ssh                :::*  users:(("sshd",1551,4))
LISTEN   0      100               ::1:smtp                :::*  users:(("master",1704,14))

2.6 The /proc File System

The /proc file system is a pseudo file system in which the kernel reserves important information in the form of virtual files. For example, display the CPU type with this command:

tux > cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 30
model name      : Intel(R) Core(TM) i5 CPU         750  @ 2.67GHz
stepping        : 5
microcode       : 0x6
cpu MHz         : 1197.000
cache size      : 8192 KB
physical id     : 0
siblings        : 4
core id         : 0
cpu cores       : 4
apicid          : 0
initial apicid  : 0
fpu             : yes
fpu_exception   : yes
cpuid level     : 11
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 sse4_2 popcnt lahf_lm ida dtherm tpr_shadow vnmi flexpriority ept vpid
bogomips        : 5333.85
clflush size    : 64
cache_alignment : 64
address sizes   : 36 bits physical, 48 bits virtual
power management:
[...]
Tip
Tip: Detailed Processor Information

Detailed information about the processor on the AMD64/Intel 64 architecture is also available by running x86info.

Query the allocation and use of interrupts with the following command:

tux > cat /proc/interrupts
           CPU0       CPU1       CPU2       CPU3
  0:        121          0          0          0   IO-APIC-edge      timer
  8:          0          0          0          1   IO-APIC-edge      rtc0
  9:          0          0          0          0   IO-APIC-fasteoi   acpi
 16:          0      11933          0          0   IO-APIC-fasteoi   ehci_hcd:+
 18:          0          0          0          0   IO-APIC-fasteoi   i801_smbus
 19:          0     117978          0          0   IO-APIC-fasteoi   ata_piix,+
 22:          0          0    3275185          0   IO-APIC-fasteoi   enp5s1
 23:     417927          0          0          0   IO-APIC-fasteoi   ehci_hcd:+
 40:    2727916          0          0          0  HPET_MSI-edge      hpet2
 41:          0    2749134          0          0  HPET_MSI-edge      hpet3
 42:          0          0    2759148          0  HPET_MSI-edge      hpet4
 43:          0          0          0    2678206  HPET_MSI-edge      hpet5
 45:          0          0          0          0   PCI-MSI-edge      aerdrv, P+
 46:          0          0          0          0   PCI-MSI-edge      PCIe PME,+
 47:          0          0          0          0   PCI-MSI-edge      PCIe PME,+
 48:          0          0          0          0   PCI-MSI-edge      PCIe PME,+
 49:          0          0          0        387   PCI-MSI-edge      snd_hda_i+
 50:     933117          0          0          0   PCI-MSI-edge      nvidia
NMI:       2102       2023       2031       1920   Non-maskable interrupts
LOC:         92         71         57         41   Local timer interrupts
SPU:          0          0          0          0   Spurious interrupts
PMI:       2102       2023       2031       1920   Performance monitoring int+
IWI:      47331      45725      52464      46775   IRQ work interrupts
RTR:          2          0          0          0   APIC ICR read retries
RES:     472911     396463     339792     323820   Rescheduling interrupts
CAL:      48389      47345      54113      50478   Function call interrupts
TLB:      28410      26804      24389      26157   TLB shootdowns
TRM:          0          0          0          0   Thermal event interrupts
THR:          0          0          0          0   Threshold APIC interrupts
MCE:          0          0          0          0   Machine check exceptions
MCP:         40         40         40         40   Machine check polls
ERR:          0
MIS:          0

The address assignment of executables and libraries is contained in the maps file:

tux > cat /proc/self/maps
08048000-0804c000 r-xp 00000000 03:03 17753      /bin/cat
0804c000-0804d000 rw-p 00004000 03:03 17753      /bin/cat
0804d000-0806e000 rw-p 0804d000 00:00 0          [heap]
b7d27000-b7d5a000 r--p 00000000 03:03 11867      /usr/lib/locale/en_GB.utf8/
b7d5a000-b7e32000 r--p 00000000 03:03 11868      /usr/lib/locale/en_GB.utf8/
b7e32000-b7e33000 rw-p b7e32000 00:00 0
b7e33000-b7f45000 r-xp 00000000 03:03 8837       /lib/libc-2.3.6.so
b7f45000-b7f46000 r--p 00112000 03:03 8837       /lib/libc-2.3.6.so
b7f46000-b7f48000 rw-p 00113000 03:03 8837       /lib/libc-2.3.6.so
b7f48000-b7f4c000 rw-p b7f48000 00:00 0
b7f52000-b7f53000 r--p 00000000 03:03 11842      /usr/lib/locale/en_GB.utf8/
[...]
b7f5b000-b7f61000 r--s 00000000 03:03 9109       /usr/lib/gconv/gconv-module
b7f61000-b7f62000 r--p 00000000 03:03 9720       /usr/lib/locale/en_GB.utf8/
b7f62000-b7f76000 r-xp 00000000 03:03 8828       /lib/ld-2.3.6.so
b7f76000-b7f78000 rw-p 00013000 03:03 8828       /lib/ld-2.3.6.so
bfd61000-bfd76000 rw-p bfd61000 00:00 0          [stack]
ffffe000-fffff000 ---p 00000000 00:00 0          [vdso]

A lot more information can be obtained from the /proc file system. Some important files and their contents are:

/proc/devices

Available devices

/proc/modules

Kernel modules loaded

/proc/cmdline

Kernel command line

/proc/meminfo

Detailed information about memory usage

/proc/config.gz

gzip-compressed configuration file of the kernel currently running

/proc/PID/

Find information about processes currently running in the /proc/NNN directories, where NNN is the process ID (PID) of the relevant process. Every process can find its own characteristics in /proc/self/.

Further information is available in the text file /usr/src/linux/Documentation/filesystems/proc.txt (this file is available when the package kernel-source is installed).

2.6.1 procinfo

Important information from the /proc file system is summarized by the command procinfo:

tux > procinfo
Linux 3.11.10-17-desktop (geeko@buildhost) (gcc 4.8.1 20130909) #1 4CPU [jupiter.example.com]

Memory:      Total        Used        Free      Shared     Buffers      Cached
Mem:       8181908     8000632      181276           0       85472     2850872
Swap:     10481660        1576    10480084

Bootup: Mon Jul 28 09:54:13 2014    Load average: 1.61 0.85 0.74 2/904 25949

user  :       1:54:41.84  12.7%  page in :    2107312  disk 1:    52212r   20199w
nice  :       0:00:00.46   0.0%  page out:    1714461  disk 2:    19387r   10928w
system:       0:25:38.00   2.8%  page act:     466673  disk 3:      548r      10w
IOwait:       0:04:16.45   0.4%  page dea:     272297
hw irq:       0:00:00.42   0.0%  page flt:  105754526
sw irq:       0:01:26.48   0.1%  swap in :          0
idle  :      12:14:43.65  81.5%  swap out:        394
guest :       0:02:18.59   0.2%
uptime:       3:45:22.24         context :   99809844

irq  0:       121 timer                 irq 41:   3238224 hpet3
irq  8:         1 rtc0                  irq 42:   3251898 hpet4
irq  9:         0 acpi                  irq 43:   3156368 hpet5
irq 16:     14589 ehci_hcd:usb1         irq 45:         0 aerdrv, PCIe PME
irq 18:         0 i801_smbus            irq 46:         0 PCIe PME, pciehp
irq 19:    124861 ata_piix, ata_piix, f irq 47:         0 PCIe PME, pciehp
irq 22:   3742817 enp5s1                irq 48:         0 PCIe PME, pciehp
irq 23:    479248 ehci_hcd:usb2         irq 49:       387 snd_hda_intel
irq 40:   3216894 hpet2                 irq 50:   1088673 nvidia

To see all the information, use the parameter -a. The parameter -nN produces updates of the information every N seconds. In this case, terminate the program by pressing Q.

By default, the cumulative values are displayed. The parameter -d produces the differential values. procinfo -dn5 displays the values that have changed in the last five seconds:

2.6.2 System Control Parameters: /proc/sys/

System control parameters are used to modify the Linux kernel parameters at runtime. They reside in /proc/sys/ and can be viewed and modified with the sysctl command. To list all parameters, run sysctl -a. A single parameter can be listed with sysctl PARAMETER_NAME.

Parameters are grouped into categories and can be listed with sysctl CATEGORY or by listing the contents of the respective directories. The most important categories are listed below. The links to further readings require the installation of the package kernel-source.

sysctl dev (/proc/sys/dev/)

Device-specific information.

sysctl fs (/proc/sys/fs/)

Used file handles, quotas, and other file system-oriented parameters. For details see /usr/src/linux/Documentation/sysctl/fs.txt.

sysctl kernel (/proc/sys/kernel/)

Information about the task scheduler, system shared memory, and other kernel-related parameters. For details see /usr/src/linux/Documentation/sysctl/kernel.txt

systctl net (/proc/sys/net/)

Information about network bridges, and general network parameters (mainly the ipv4/ subdirectory). For details see /usr/src/linux/Documentation/sysctl/net.txt

sysctl vm (/proc/sys/vm/)

Entries in this path relate to information about the virtual memory, swapping, and caching. For details see /usr/src/linux/Documentation/sysctl/vm.txt

To set or change a parameter for the current session, use the command sysctl -w PARAMETER=VALUE. To permanently change a setting, add a line PARAMETER=VALUE to /etc/sysctl.conf.

2.7 Hardware Information

2.7.1 PCI Resources: lspci

Note
Note: Accessing PCI configuration.

Most operating systems require root user privileges to grant access to the computer's PCI configuration.

The command lspci lists the PCI resources:

root # lspci
00:00.0 Host bridge: Intel Corporation 82845G/GL[Brookdale-G]/GE/PE \
    DRAM Controller/Host-Hub Interface (rev 01)
00:01.0 PCI bridge: Intel Corporation 82845G/GL[Brookdale-G]/GE/PE \
    Host-to-AGP Bridge (rev 01)
00:1d.0 USB Controller: Intel Corporation 82801DB/DBL/DBM \
    (ICH4/ICH4-L/ICH4-M) USB UHCI Controller #1 (rev 01)
00:1d.1 USB Controller: Intel Corporation 82801DB/DBL/DBM \
    (ICH4/ICH4-L/ICH4-M) USB UHCI Controller #2 (rev 01)
00:1d.2 USB Controller: Intel Corporation 82801DB/DBL/DBM \
    (ICH4/ICH4-L/ICH4-M) USB UHCI Controller #3 (rev 01)
00:1d.7 USB Controller: Intel Corporation 82801DB/DBM \
    (ICH4/ICH4-M) USB2 EHCI Controller (rev 01)
00:1e.0 PCI bridge: Intel Corporation 82801 PCI Bridge (rev 81)
00:1f.0 ISA bridge: Intel Corporation 82801DB/DBL (ICH4/ICH4-L) \
    LPC Interface Bridge (rev 01)
00:1f.1 IDE interface: Intel Corporation 82801DB (ICH4) IDE \
    Controller (rev 01)
00:1f.3 SMBus: Intel Corporation 82801DB/DBL/DBM (ICH4/ICH4-L/ICH4-M) \
    SMBus Controller (rev 01)
00:1f.5 Multimedia audio controller: Intel Corporation 82801DB/DBL/DBM \
    (ICH4/ICH4-L/ICH4-M) AC'97 Audio Controller (rev 01)
01:00.0 VGA compatible controller: Matrox Graphics, Inc. G400/G450 (rev 85)
02:08.0 Ethernet controller: Intel Corporation 82801DB PRO/100 VE (LOM) \
    Ethernet Controller (rev 81)

Using -v results in a more detailed listing:

root # lspci -v
[...]
00:03.0 Ethernet controller: Intel Corporation 82540EM Gigabit Ethernet \
Controller (rev 02)
  Subsystem: Intel Corporation PRO/1000 MT Desktop Adapter
  Flags: bus master, 66MHz, medium devsel, latency 64, IRQ 19
  Memory at f0000000 (32-bit, non-prefetchable) [size=128K]
  I/O ports at d010 [size=8]
  Capabilities: [dc] Power Management version 2
  Capabilities: [e4] PCI-X non-bridge device
  Kernel driver in use: e1000
  Kernel modules: e1000

Information about device name resolution is obtained from the file /usr/share/pci.ids. PCI IDs not listed in this file are marked Unknown device.

The parameter -vv produces all the information that could be queried by the program. To view the pure numeric values, use the parameter -n.

2.7.2 USB Devices: lsusb

The command lsusb lists all USB devices. With the option -v, print a more detailed list. The detailed information is read from the directory /proc/bus/usb/. The following is the output of lsusb with these USB devices attached: hub, memory stick, hard disk and mouse.

root # lsusb
Bus 004 Device 007: ID 0ea0:2168 Ours Technology, Inc. Transcend JetFlash \
    2.0 / Astone USB Drive
Bus 004 Device 006: ID 04b4:6830 Cypress Semiconductor Corp. USB-2.0 IDE \
    Adapter
Bus 004 Device 005: ID 05e3:0605 Genesys Logic, Inc.
Bus 004 Device 001: ID 0000:0000
Bus 003 Device 001: ID 0000:0000
Bus 002 Device 001: ID 0000:0000
Bus 001 Device 005: ID 046d:c012 Logitech, Inc. Optical Mouse
Bus 001 Device 001: ID 0000:0000

2.7.3 Monitoring and Tuning the Thermal Subsystem: tmon

tmon is a tool to help visualize, tune, and test the complex thermal subsystem. When started without parameters, tmon runs in monitoring mode:

┌──────THERMAL ZONES(SENSORS)──────────────────────────────┐
│Thermal Zones:                 acpitz00                   │
│Trip Points:                   PC                         │
└──────────────────────────────────────────────────────────┘
┌─────────── COOLING DEVICES ──────────────────────────────┐
│ID  Cooling Dev   Cur    Max   Thermal Zone Binding       │
│00    Processor     0      3   ││││││││││││               │
│01    Processor     0      3   ││││││││││││               │
│02    Processor     0      3   ││││││││││││               │
│03    Processor     0      3   ││││││││││││               │
│04 intel_powerc    -1     50   ││││││││││││               │
└──────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────┐
│                         10        20        30        40 │
│acpitz 0:[  8][>>>>>>>>>P9                    C31         │
└──────────────────────────────────────────────────────────┘
┌────────────────── CONTROLS ──────────────────────────────┐
│PID gain: kp=0.36 ki=5.00 kd=0.19 Output 0.00             │
│Target Temp: 65.0C, Zone: 0, Control Device: None         │
└──────────────────────────────────────────────────────────┘

 Ctrl-c - Quit   TAB - Tuning

For detailed information on how to interpret the data, how to log thermal data and how to use tmon to test and tune cooling devices and sensors, refer to the man page: man 8 tmon. The package tmon is not installed by default.

2.7.4 MCELog: Machine Check Exceptions (MCE)

The mcelog package logs and parses/translates Machine Check Exceptions (MCE) on hardware errors (also including memory errors). Formerly this has been done by a cron job executed hourly. Now hardware errors are immediately processed by an mcelog daemon.

However, the mcelog service is not enabled by default, resulting in memory and CPU errors also not being logged by default. In addition, mcelog has a new feature to also handle predictive bad page offlining and automatic core offlining when cache errors happen.

The service can either be enabled and started via the YaST system services editor or via command line:

root # systemctl enable mcelog
root # systemctl start mcelog

2.7.5 x86_64: dmidecode: DMI Table Decoder

dmidecode shows the machine's DMI table containing information such as serial numbers and BIOS revisions of the hardware.

root # dmidecode
# dmidecode 2.12
SMBIOS 2.5 present.
27 structures occupying 1298 bytes.
Table at 0x000EB250.

Handle 0x0000, DMI type 4, 35 bytes
Processor Information
        Socket Designation: J1PR
        Type: Central Processor
        Family: Other
        Manufacturer: Intel(R) Corporation
        ID: E5 06 01 00 FF FB EB BF
        Version: Intel(R) Core(TM) i5 CPU         750  @ 2.67GHz
        Voltage: 1.1 V
        External Clock: 133 MHz
        Max Speed: 4000 MHz
        Current Speed: 2667 MHz
        Status: Populated, Enabled
        Upgrade: Other
        L1 Cache Handle: 0x0004
        L2 Cache Handle: 0x0003
        L3 Cache Handle: 0x0001
        Serial Number: Not Specified
        Asset Tag: Not Specified
        Part Number: Not Specified
[..]

2.8 Files and File Systems

For file system-specific information, refer to Book “Storage Administration Guide”.

2.8.1 Determine the File Type: file

The command file determines the type of a file or a list of files by checking /usr/share/misc/magic.

tux > file /usr/bin/file
/usr/bin/file: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), \
    for GNU/Linux 2.6.4, dynamically linked (uses shared libs), stripped

The parameter -f LIST specifies a file with a list of file names to examine. The -z allows file to look inside compressed files:

tux > file /usr/share/man/man1/file.1.gz
/usr/share/man/man1/file.1.gz: gzip compressed data, from Unix, max compression
tux > file -z /usr/share/man/man1/file.1.gz
/usr/share/man/man1/file.1.gz: troff or preprocessor input text \
    (gzip compressed data, from Unix, max compression)

The parameter -i outputs a mime type string rather than the traditional description.

tux > file -i /usr/share/misc/magic
/usr/share/misc/magic: text/plain charset=utf-8

2.8.2 File Systems and Their Usage: mount, df and du

The command mount shows which file system (device and type) is mounted at which mount point:

root # mount
/dev/sda2 on / type ext4 (rw,acl,user_xattr)
proc on /proc type proc (rw)
sysfs on /sys type sysfs (rw)
debugfs on /sys/kernel/debug type debugfs (rw)
devtmpfs on /dev type devtmpfs (rw,mode=0755)
tmpfs on /dev/shm type tmpfs (rw,mode=1777)
devpts on /dev/pts type devpts (rw,mode=0620,gid=5)
/dev/sda3 on /home type ext3 (rw)
securityfs on /sys/kernel/security type securityfs (rw)
fusectl on /sys/fs/fuse/connections type fusectl (rw)
gvfs-fuse-daemon on /home/tux/.gvfs type fuse.gvfs-fuse-daemon \
(rw,nosuid,nodev,user=tux)

Obtain information about total usage of the file systems with the command df. The parameter -h (or --human-readable) transforms the output into a form understandable for common users.

tux > df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/sda2              20G  5,9G   13G  32% /
devtmpfs              1,6G  236K  1,6G   1% /dev
tmpfs                 1,6G  668K  1,6G   1% /dev/shm
/dev/sda3             208G   40G  159G  20% /home

Display the total size of all the files in a given directory and its subdirectories with the command du. The parameter -s suppresses the output of detailed information and gives only a total for each argument. -h again transforms the output into a human-readable form:

tux > du -sh /opt
192M    /opt

2.8.3 Additional Information about ELF Binaries

Read the content of binaries with the readelf utility. This even works with ELF files that were built for other hardware architectures:

tux > readelf --file-header /bin/ls
ELF Header:
  Magic:   7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF64
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           Advanced Micro Devices X86-64
  Version:                           0x1
  Entry point address:               0x402540
  Start of program headers:          64 (bytes into file)
  Start of section headers:          95720 (bytes into file)
  Flags:                             0x0
  Size of this header:               64 (bytes)
  Size of program headers:           56 (bytes)
  Number of program headers:         9
  Size of section headers:           64 (bytes)
  Number of section headers:         32
  Section header string table index: 31

2.8.4 File Properties: stat

The command stat displays file properties:

tux > stat /etc/profile
  File: `/etc/profile'
  Size: 9662            Blocks: 24         IO Block: 4096   regular file
Device: 802h/2050d      Inode: 132349      Links: 1
Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2009-03-20 07:51:17.000000000 +0100
Modify: 2009-01-08 19:21:14.000000000 +0100
Change: 2009-03-18 12:55:31.000000000 +0100

The parameter --file-system produces details of the properties of the file system in which the specified file is located:

tux > stat /etc/profile --file-system
  File: "/etc/profile"
    ID: d4fb76e70b4d1746 Namelen: 255     Type: ext2/ext3
Block size: 4096       Fundamental block size: 4096
Blocks: Total: 2581445    Free: 1717327    Available: 1586197
Inodes: Total: 655776     Free: 490312

2.9 User Information

2.9.1 User Accessing Files: fuser

It can be useful to determine what processes or users are currently accessing certain files. Suppose, for example, you want to unmount a file system mounted at /mnt. umount returns "device is busy." The command fuser can then be used to determine what processes are accessing the device:

tux > fuser -v /mnt/*

                     USER        PID ACCESS COMMAND
/mnt/notes.txt       tux    26597 f....  less

Following termination of the less process, which was running on another terminal, the file system can successfully be unmounted. When used with -k option, fuser will terminate processes accessing the file as well.

2.9.2 Who Is Doing What: w

With the command w, find out who is logged in to the system and what each user is doing. For example:

tux > w
 16:00:59 up 1 day,  2:41,  3 users,  load average: 0.00, 0.01, 0.05
USER     TTY      FROM             LOGIN@   IDLE   JCPU   PCPU WHAT
tux      :0       console          Wed13   ?xdm?   8:15   0.03s /usr/lib/gdm/gd
tux      console  :0               Wed13   26:41m  0.00s  0.03s /usr/lib/gdm/gd
tux      pts/0    :0               Wed13   20:11   0.10s  2.89s /usr/lib/gnome-

If any users of other systems have logged in remotely, the parameter -f shows the computers from which they have established the connection.

2.10 Time and Date

2.10.1 Time Measurement with time

Determine the time spent by commands with the time utility. This utility is available in two versions: as a Bash built-in and as a program (/usr/bin/time).

tux > time find . > /dev/null

real    0m4.051s1
user    0m0.042s2
sys     0m0.205s3

1

The real time that elapsed from the command's start-up until it finished.

2

CPU time of the user as reported by the times system call.

3

CPU time of the system as reported by the times system call.

The output of /usr/bin/time is much more detailed. It is recommended to run it with the -v switch to produce human-readable output.

/usr/bin/time -v find . > /dev/null
        Command being timed: "find ."
        User time (seconds): 0.24
        System time (seconds): 2.08
        Percent of CPU this job got: 25%
        Elapsed (wall clock) time (h:mm:ss or m:ss): 0:09.03
        Average shared text size (kbytes): 0
        Average unshared data size (kbytes): 0
        Average stack size (kbytes): 0
        Average total size (kbytes): 0
        Maximum resident set size (kbytes): 2516
        Average resident set size (kbytes): 0
        Major (requiring I/O) page faults: 0
        Minor (reclaiming a frame) page faults: 1564
        Voluntary context switches: 36660
        Involuntary context switches: 496
        Swaps: 0
        File system inputs: 0
        File system outputs: 0
        Socket messages sent: 0
        Socket messages received: 0
        Signals delivered: 0
        Page size (bytes): 4096
        Exit status: 0

2.11 Graph Your Data: RRDtool

There are a lot of data in the world around you, which can be easily measured in time. For example, changes in the temperature, or the number of data sent or received by your computer's network interface. RRDtool can help you store and visualize such data in detailed and customizable graphs.

RRDtool is available for most Unix platforms and Linux distributions. SUSE® Linux Enterprise Server ships RRDtool as well. Install it either with YaST or by entering

zypper install rrdtool in the command line as root.

Tip
Tip: Bindings

There are Perl, Python, Ruby, and PHP bindings available for RRDtool, so that you can write your own monitoring scripts in your preferred scripting language.

2.11.1 How RRDtool Works

RRDtool is an abbreviation of Round Robin Database tool. Round Robin is a method for manipulating with a constant amount of data. It uses the principle of a circular buffer, where there is no end nor beginning to the data row which is being read. RRDtool uses Round Robin Databases to store and read its data.

As mentioned above, RRDtool is designed to work with data that change in time. The ideal case is a sensor which repeatedly reads measured data (like temperature, speed etc.) in constant periods of time, and then exports them in a given format. Such data are perfectly ready for RRDtool, and it is easy to process them and create the desired output.

Sometimes it is not possible to obtain the data automatically and regularly. Their format needs to be pre-processed before it is supplied to RRDtool, and often you need to manipulate RRDtool even manually.

The following is a simple example of basic RRDtool usage. It illustrates all three important phases of the usual RRDtool workflow: creating a database, updating measured values, and viewing the output.

2.11.2 A Practical Example

Suppose we want to collect and view information about the memory usage in the Linux system as it changes in time. To make the example more vivid, we measure the currently free memory over a period of 40 seconds in 4-second intervals. Three applications that usually consume a lot of system memory are started and closed: the Firefox Web browser, the Evolution e-mail client, and the Eclipse development framework.

2.11.2.1 Collecting Data

RRDtool is very often used to measure and visualize network traffic. In such case, the Simple Network Management Protocol (SNMP) is used. This protocol can query network devices for relevant values of their internal counters. Exactly these values are to be stored with RRDtool. For more information on SNMP, see http://www.net-snmp.org/.

Our situation is different—we need to obtain the data manually. A helper script free_mem.sh repetitively reads the current state of free memory and writes it to the standard output.

tux > cat free_mem.sh
INTERVAL=4
for steps in {1..10}
do
    DATE=`date +%s`
    FREEMEM=`free -b | grep "Mem" | awk '{ print $4 }'`
    sleep $INTERVAL
    echo "rrdtool update free_mem.rrd $DATE:$FREEMEM"
done
  • The time interval is set to 4 seconds, and is implemented with the sleep command.

  • RRDtool accepts time information in a special format - so called Unix time. It is defined as the number of seconds since the midnight of January 1, 1970 (UTC). For example, 1272907114 represents 2010-05-03 17:18:34.

  • The free memory information is reported in bytes with free -b. Prefer to supply basic units (bytes) instead of multiple units (like kilobytes).

  • The line with the echo ... command contains the future name of the database file (free_mem.rrd), and together creates a command line for updating RRDtool values.

After running free_mem.sh, you see an output similar to this:

tux > sh free_mem.sh
rrdtool update free_mem.rrd 1272974835:1182994432
rrdtool update free_mem.rrd 1272974839:1162817536
rrdtool update free_mem.rrd 1272974843:1096269824
rrdtool update free_mem.rrd 1272974847:1034219520
rrdtool update free_mem.rrd 1272974851:909438976
rrdtool update free_mem.rrd 1272974855:832454656
rrdtool update free_mem.rrd 1272974859:829120512
rrdtool update free_mem.rrd 1272974863:1180377088
rrdtool update free_mem.rrd 1272974867:1179369472
rrdtool update free_mem.rrd 1272974871:1181806592

It is convenient to redirect the command's output to a file with

sh free_mem.sh > free_mem_updates.log

to simplify its future execution.

2.11.2.2 Creating the Database

Create the initial Robin Round database for our example with the following command:

tux > rrdtool create free_mem.rrd --start 1272974834 --step=4 \
DS:memory:GAUGE:600:U:U RRA:AVERAGE:0.5:1:24
Points to Notice
  • This command creates a file called free_mem.rrd for storing our measured values in a Round Robin type database.

  • The --start option specifies the time (in Unix time) when the first value will be added to the database. In this example, it is one less than the first time value of the free_mem.sh output (1272974835).

  • The --step specifies the time interval in seconds with which the measured data will be supplied to the database.

  • The DS:memory:GAUGE:600:U:U part introduces a new data source for the database. It is called memory, its type is gauge, the maximum number between two updates is 600 seconds, and the minimal and maximal value in the measured range are unknown (U).

  • RRA:AVERAGE:0.5:1:24 creates Round Robin archive (RRA) whose stored data are processed with the consolidation functions (CF) that calculates the average of data points. 3 arguments of the consolidation function are appended to the end of the line.

If no error message is displayed, then free_mem.rrd database is created in the current directory:

tux > ls -l free_mem.rrd
-rw-r--r-- 1 tux users 776 May  5 12:50 free_mem.rrd

2.11.2.3 Updating Database Values

After the database is created, you need to fill it with the measured data. In Section 2.11.2.1, “Collecting Data”, we already prepared the file free_mem_updates.log which consists of rrdtool update commands. These commands do the update of database values for us.

tux > sh free_mem_updates.log; ls -l free_mem.rrd
-rw-r--r--  1 tux users  776 May  5 13:29 free_mem.rrd

As you can see, the size of free_mem.rrd remained the same even after updating its data.

2.11.2.4 Viewing Measured Values

We have already measured the values, created the database, and stored the measured value in it. Now we can play with the database, and retrieve or view its values.

To retrieve all the values from our database, enter the following on the command line:

tux > rrdtool fetch free_mem.rrd AVERAGE --start 1272974830 \
--end 1272974871
          memory
1272974832: nan
1272974836: 1.1729059840e+09
1272974840: 1.1461806080e+09
1272974844: 1.0807572480e+09
1272974848: 1.0030243840e+09
1272974852: 8.9019289600e+08
1272974856: 8.3162112000e+08
1272974860: 9.1693465600e+08
1272974864: 1.1801251840e+09
1272974868: 1.1799787520e+09
1272974872: nan
Points to Notice
  • AVERAGE will fetch average value points from the database, because only one data source is defined (Section 2.11.2.2, “Creating the Database”) with AVERAGE processing and no other function is available.

  • The first line of the output prints the name of the data source as defined in Section 2.11.2.2, “Creating the Database”.

  • The left results column represents individual points in time, while the right one represents corresponding measured average values in scientific notation.

  • The nan in the last line stands for not a number.

Now a graph representing the values stored in the database is drawn:

tux > rrdtool graph free_mem.png \
--start 1272974830 \
--end 1272974871 \
--step=4 \
DEF:free_memory=free_mem.rrd:memory:AVERAGE \
LINE2:free_memory#FF0000 \
--vertical-label "GB" \
--title "Free System Memory in Time" \
--zoom 1.5 \
--x-grid SECOND:1:SECOND:4:SECOND:10:0:%X
Points to Notice
  • free_mem.png is the file name of the graph to be created.

  • --start and --end limit the time range within which the graph will be drawn.

  • --step specifies the time resolution (in seconds) of the graph.

  • The DEF:... part is a data definition called free_memory. Its data are read from the free_mem.rrd database and its data source called memory. The average value points are calculated, because no others were defined in Section 2.11.2.2, “Creating the Database”.

  • The LINE... part specifies properties of the line to be drawn into the graph. It is 2 pixels wide, its data come from the free_memory definition, and its color is red.

  • --vertical-label sets the label to be printed along the y axis, and --title sets the main label for the whole graph.

  • --zoom specifies the zoom factor for the graph. This value must be greater than zero.

  • --x-grid specifies how to draw grid lines and their labels into the graph. Our example places them every second, while major grid lines are placed every 4 seconds. Labels are placed every 10 seconds under the major grid lines.

Example Graph Created with RRDtool
Figure 2.1: Example Graph Created with RRDtool

2.11.3 For More Information

RRDtool is a very complex tool with a lot of sub-commands and command line options. Some are easy to understand, but to make it produce the results you want and fine-tune them according to your liking may require a lot of effort.

Apart from RRDtool's man page (man 1 rrdtool) which gives you only basic information, you should have a look at the RRDtool home page. There is a detailed documentation of the rrdtool command and all its sub-commands. There are also several tutorials to help you understand the common RRDtool workflow.

If you are interested in monitoring network traffic, have a look at MRTG (Multi Router Traffic Grapher). MRTG can graph the activity of many network devices. It can use RRDtool.

3 Analyzing and Managing System Log Files

System log file analysis is one of the most important tasks when analyzing the system. In fact, looking at the system log files should be the first thing to do when maintaining or troubleshooting a system. SUSE Linux Enterprise Server automatically logs almost everything that happens on the system in detail. Since the move to systemd, kernel messages and messages of system services registered with systemd are logged in systemd journal (see Book “Administration Guide”, Chapter 16 “journalctl: Query the systemd Journal”). Other log files (mainly those of system applications) are written in plain text and can be easily read using an editor or pager. It is also possible to parse them using scripts. This allows you to filter their content.

3.1 System Log Files in /var/log/

System log files are always located under the /var/log directory. The following list presents an overview of all system log files from SUSE Linux Enterprise Server present after a default installation. Depending on your installation scope, /var/log also contains log files from other services and applications not listed here. Some files and directories described below are placeholders and are only used, when the corresponding application is installed. Most log files are only visible for the user root.

apparmor/

AppArmor log files. See Book “Security and Hardening Guide” for details of AppArmor.

audit/

Logs from the audit framework. See Book “Security and Hardening Guide” for details.

ConsoleKit/

Logs of the ConsoleKit daemon (daemon for tracking what users are logged in and how they interact with the computer).

cups/

Access and error logs of the Common Unix Printing System (cups).

firewall

Firewall logs.

gdm/

Log files from the GNOME display manager.

krb5/

Log files from the Kerberos network authentication system.

lastlog

A database containing information on the last login of each user. Use the command lastlog to view. See man 8 lastlog for more information.

localmessages

Log messages of some boot scripts, for example the log of the DHCP client.

mail*

Mail server (postfix, sendmail) logs.

messages

This is the default place where all kernel and system log messages go and should be the first place (along with /var/log/warn) to look at in case of problems.

NetworkManager

NetworkManager log files.

news/

Log messages from a news server.

ntp

Logs from the Network Time Protocol daemon (ntpd).

pk_backend_zypp*

PackageKit (with libzypp back-end) log files.

puppet/

Log files from the data center automation tool puppet.

samba/

Log files from Samba, the Windows SMB/CIFS file server.

warn

Log of all system warnings and errors. This should be the first place (along with the output of the systemd journal) to look in case of problems.

wtmp

Database of all login/logout activities, and remote connections. Use the command last to view. See man 1 last for more information.

xinetd.log

Log files from the extended Internet services daemon (xinetd).

Xorg.0.log

X.Org start-up log file. Refer to this in case you have problems starting X.Org. Copies from previous X.Org starts are numbered Xorg.?.log.

YaST2/

All YaST log files.

zypp/

libzypp log files. Refer to these files for the package installation history.

zypper.log

Logs from the command line installer zypper.

3.2 Viewing and Parsing Log Files

To view log files, you can use any text editor. There is also a simple YaST module for viewing the system log available in the YaST control center under Miscellaneous › System Log.

For viewing log files in a text console, use the commands less or more. Use head and tail to view the beginning or end of a log file. To view entries appended to a log file in real-time use tail -f. For information about how to use these tools, see their man pages.

To search for strings or regular expressions in log files use grep. awk is useful for parsing and rewriting log files.

3.3 Managing Log Files with logrotate

Log files under /var/log grow on a daily basis and quickly become very large. logrotate is a tool that helps you manage log files and their growth. It allows automatic rotation, removal, compression, and mailing of log files. Log files can be handled periodically (daily, weekly, or monthly) or when exceeding a particular size.

logrotate is usually run daily by systemd, and thus usually modifies log files only once a day. However, exceptions occur when a log file is modified because of its size, if logrotate is run multiple times a day, or if --force is enabled. Use /var/lib/misc/logrotate.status to find out when a particular file was last rotated.

The main configuration file of logrotate is /etc/logrotate.conf. System packages and programs that produce log files (for example, apache2) put their own configuration files in the /etc/logrotate.d/ directory. The content of /etc/logrotate.d/ is included via /etc/logrotate.conf.

Example 3.1: Example for /etc/logrotate.conf
# see "man logrotate" for details
# rotate log files weekly
weekly

# keep 4 weeks worth of backlogs
rotate 4

# create new (empty) log files after rotating old ones
create

# use date as a suffix of the rotated file
dateext

# uncomment this if you want your log files compressed
#compress

# comment these to switch compression to use gzip or another
# compression scheme
compresscmd /usr/bin/bzip2
uncompresscmd /usr/bin/bunzip2

# RPM packages drop log rotation information into this directory
include /etc/logrotate.d
Important
Important: Avoid Permission Conflicts

The create option pays heed to the modes and ownerships of files specified in /etc/permissions*. If you modify these settings, make sure no conflicts arise.

3.4 Monitoring Log Files with logwatch

logwatch is a customizable, pluggable log-monitoring script. It parses system logs, extracts the important information and presents them in a human readable manner. To use logwatch, install the logwatch package.

logwatch can either be used at the command line to generate on-the-fly reports, or via cron to regularly create custom reports. Reports can either be printed on the screen, saved to a file, or be mailed to a specified address. The latter is especially useful when automatically generating reports via cron.

On the command line, you can tell logwatch for which service and time span to generate a report and how much detail should be included:

# Detailed report on all kernel messages from yesterday
logwatch --service kernel --detail High --range Yesterday --print

# Low detail report on all sshd events recorded (incl. archived logs)
logwatch --service sshd --detail Low --range All --archives --print

# Mail a report on all smartd messages from May 5th to May 7th to root@localhost
logwatch --service smartd --range 'between 5/5/2005 and 5/7/2005' \
--mailto root@localhost --print

The --range option has got a complex syntax—see logwatch --range help for details. A list of all services that can be queried is available with the following command:

ls /usr/share/logwatch/default.conf/services/ | sed 's/\.conf//g'

logwatch can be customized to great detail. However, the default configuration should usually be sufficient. The default configuration files are located under /usr/share/logwatch/default.conf/. Never change them because they would get overwritten again with the next update. Rather place custom configuration in /etc/logwatch/conf/ (you may use the default configuration file as a template, though). A detailed HOWTO on customizing logwatch is available at /usr/share/doc/packages/logwatch/HOWTO-Customize-LogWatch. The following configuration files exist:

logwatch.conf

The main configuration file. The default version is extensively commented. Each configuration option can be overwritten on the command line.

ignore.conf

Filter for all lines that should globally be ignored by logwatch.

services/*.conf

The service directory holds configuration files for each service you can generate a report for.

logfiles/*.conf

Specifications on which log files should be parsed for each service.

3.5 Using logger to Make System Log Entries

logger is a tool for making entries in the system log. It provides a shell command interface to the rsyslogd system log module. For example, the following line outputs its message in /var/log/messages or directly in the journal (if no logging facility is running):

logger -t Test "This message comes from $USER"

Depending on the current user and host name, the log contains a line similar to this:

Sep 28 13:09:31 venus Test: This message comes from tux

Part III Kernel Monitoring

  • 4 SystemTap—Filtering and Analyzing System Data
  • SystemTap provides a command line interface and a scripting language to examine the activities of a running Linux system, particularly the kernel, in fine detail. SystemTap scripts are written in the SystemTap scripting language, are then compiled to C-code kernel modules and inserted into the kerne…

  • 5 Kernel Probes
  • Kernel probes are a set of tools to collect Linux kernel debugging and performance information. Developers and system administrators usually use them either to debug the kernel, or to find system performance bottlenecks. The reported data can then be used to tune the system for better performance.

  • 6 Hardware-Based Performance Monitoring with Perf
  • Perf is an interface to access the performance monitoring unit (PMU) of a processor and to record and display software events such as page faults. It supports system-wide, per-thread, and KVM virtualization guest monitoring.

  • 7 OProfile—System-Wide Profiler
  • OProfile is a profiler for dynamic program analysis. It investigates the behavior of a running program and gathers information. This information can be viewed and gives hints for further optimization.

    It is not necessary to recompile or use wrapper libraries to use OProfile. Not even a kernel patch is needed. Usually, when profiling an application, a small overhead is expected, depending on the workload and sampling frequency.

4 SystemTap—Filtering and Analyzing System Data

SystemTap provides a command line interface and a scripting language to examine the activities of a running Linux system, particularly the kernel, in fine detail. SystemTap scripts are written in the SystemTap scripting language, are then compiled to C-code kernel modules and inserted into the kernel. The scripts can be designed to extract, filter and summarize data, thus allowing the diagnosis of complex performance problems or functional problems. SystemTap provides information similar to the output of tools like netstat, ps, top, and iostat. However, more filtering and analysis options can be used for the collected information.

4.1 Conceptual Overview

Each time you run a SystemTap script, a SystemTap session is started. Several passes are done on the script before it is allowed to run. Then, the script is compiled into a kernel module and loaded. If the script has been executed before and no system components have changed (for example, different compiler or kernel versions, library paths, or script contents), SystemTap does not compile the script again. Instead, it uses the *.c and *.ko data stored in the SystemTap cache (~/.systemtap).

The module is unloaded when the tap has finished running. For an example, see the test run in Section 4.2, “Installation and Setup” and the respective explanation.

4.1.1 SystemTap Scripts

SystemTap usage is based on SystemTap scripts (*.stp). They tell SystemTap which type of information to collect, and what to do once that information is collected. The scripts are written in the SystemTap scripting language that is similar to AWK and C. For the language definition, see https://sourceware.org/systemtap/langref/. A lot of useful example scripts are available from http://www.sourceware.org/systemtap/examples/.

The essential idea behind a SystemTap script is to name events, and to give them handlers. When SystemTap runs the script, it monitors for certain events. When an event occurs, the Linux kernel runs the handler as a sub-routine, then resumes. Thus, events serve as the triggers for handlers to run. Handlers can record specified data and print it in a certain manner.

The SystemTap language only uses a few data types (integers, strings, and associative arrays of these), and full control structures (blocks, conditionals, loops, functions). It has a lightweight punctuation (semicolons are optional) and does not need detailed declarations (types are inferred and checked automatically).

For more information about SystemTap scripts and their syntax, refer to Section 4.3, “Script Syntax” and to the stapprobes and stapfuncs man pages, that are available with the systemtap-docs package.

4.1.2 Tapsets

Tapsets are a library of pre-written probes and functions that can be used in SystemTap scripts. When a user runs a SystemTap script, SystemTap checks the script's probe events and handlers against the tapset library. SystemTap then loads the corresponding probes and functions before translating the script to C. Like SystemTap scripts themselves, tapsets use the file name extension *.stp.

However, unlike SystemTap scripts, tapsets are not meant for direct execution. They constitute the library from which other scripts can pull definitions. Thus, the tapset library is an abstraction layer designed to make it easier for users to define events and functions. Tapsets provide aliases for functions that users could want to specify as an event. Knowing the proper alias is often easier than remembering specific kernel functions that might vary between kernel versions.

4.1.3 Commands and Privileges

The main commands associated with SystemTap are stap and staprun. To execute them, you either need root privileges or must be a member of the stapdev or stapusr group.

stap

SystemTap front-end. Runs a SystemTap script (either from file, or from standard input). It translates the script into C code, compiles it, and loads the resulting kernel module into a running Linux kernel. Then, the requested system trace or probe functions are performed.

staprun

SystemTap back-end. Loads and unloads kernel modules produced by the SystemTap front-end.

For a list of options for each command, use --help. For details, refer to the stap and the staprun man pages.

To avoid giving root access to users solely to enable them to work with SystemTap, use one of the following SystemTap groups. They are not available by default on SUSE Linux Enterprise Server, but you can create the groups and modify the access rights accordingly. Also adjust the permissions of the staprun command if the security implications are appropriate for your environment.

stapdev

Members of this group can run SystemTap scripts with stap, or run SystemTap instrumentation modules with staprun. As running stap involves compiling scripts into kernel modules and loading them into the kernel, members of this group still have effective root access.

stapusr

Members of this group are only allowed to run SystemTap instrumentation modules with staprun. In addition, they can only run those modules from /lib/modules/KERNEL_VERSION/systemtap/. This directory must be owned by root and must only be writable for the root user.

4.1.4 Important Files and Directories

The following list gives an overview of the SystemTap main files and directories.

/lib/modules/KERNEL_VERSION/systemtap/

Holds the SystemTap instrumentation modules.

/usr/share/systemtap/tapset/

Holds the standard library of tapsets.

/usr/share/doc/packages/systemtap/examples

Holds several example SystemTap scripts for various purposes. Only available if the systemtap-docs package is installed.

~/.systemtap/cache

Data directory for cached SystemTap files.

/tmp/stap*

Temporary directory for SystemTap files, including translated C code and kernel object.

4.2 Installation and Setup

As SystemTap needs information about the kernel, some additional kernel-related packages must be installed. For each kernel you want to probe with SystemTap, you need to install a set of the following packages. This set should exactly match the kernel version and flavor (indicated by * in the overview below).

Important
Important: Repository for Packages with Debugging Information

If you subscribed your system for online updates, you can find debuginfo packages in the *-Debuginfo-Updates online installation repository relevant for SUSE Linux Enterprise Server 12 SP5. Use YaST to enable the repository.

For the classic SystemTap setup, install the following packages (using either YaST or zypper).

  • systemtap

  • systemtap-server

  • systemtap-docs (optional)

  • kernel-*-base

  • kernel-*-debuginfo

  • kernel-*-devel

  • kernel-source-*

  • gcc

To get access to the man pages and to a helpful collection of example SystemTap scripts for various purposes, additionally install the systemtap-docs package.

To check if all packages are correctly installed on the machine and if SystemTap is ready to use, execute the following command as root.

stap -v -e 'probe vfs.read {printf("read performed\n"); exit()}'

It probes the currently used kernel by running a script and returning an output. If the output is similar to the following, SystemTap is successfully deployed and ready to use:

Pass 1: parsed user script and 59 library script(s) in 80usr/0sys/214real ms.
Pass 2: analyzed script: 1 probe(s), 11 function(s), 2 embed(s), 1 global(s) in
 140usr/20sys/412real ms.
Pass 3: translated to C into
 "/tmp/stapDwEk76/stap_1856e21ea1c246da85ad8c66b4338349_4970.c" in 160usr/0sys/408real ms.
Pass 4: compiled C into "stap_1856e21ea1c246da85ad8c66b4338349_4970.ko" in
 2030usr/360sys/10182real ms.
Pass 5: starting run.
 read performed
Pass 5: run completed in 10usr/20sys/257real ms.

1

Checks the script against the existing tapset library in /usr/share/systemtap/tapset/ for any tapsets used. Tapsets are scripts that form a library of pre-written probes and functions that can be used in SystemTap scripts.

2

Examines the script for its components.

3

Translates the script to C. Runs the system C compiler to create a kernel module from it. Both the resulting C code (*.c) and the kernel module (*.ko) are stored in the SystemTap cache, ~/.systemtap.

4

Loads the module and enables all the probes (events and handlers) in the script by hooking into the kernel. The event being probed is a Virtual File System (VFS) read. As the event occurs on any processor, a valid handler is executed (prints the text read performed) and closed with no errors.

5

After the SystemTap session is terminated, the probes are disabled, and the kernel module is unloaded.

In case any error messages appear during the test, check the output for hints about any missing packages and make sure they are installed correctly. Rebooting and loading the appropriate kernel may also be needed.

4.3 Script Syntax

SystemTap scripts consist of the following two components:

SystemTap Events (Probe Points)

Name the kernel events at the associated handler should be executed. Examples for events are entering or exiting a certain function, a timer expiring, or starting or terminating a session.

SystemTap Handlers (Probe Body)

Series of script language statements that specify the work to be done whenever a certain event occurs. This normally includes extracting data from the event context, storing them into internal variables, or printing results.

An event and its corresponding handler is collectively called a probe. SystemTap events are also called probe points. A probe's handler is also called a probe body.

Comments can be inserted anywhere in the SystemTap script in various styles: using either #, /* */, or // as marker.

4.3.1 Probe Format

A SystemTap script can have multiple probes. They must be written in the following format:

probe EVENT {STATEMENTS}

Each probe has a corresponding statement block. This statement block must be enclosed in { } and contains the statements to be executed per event.

Example 4.1: Simple SystemTap Script

The following example shows a simple SystemTap script.

probe1 begin2
{3
   printf4 ("hello world\n")5
   exit ()6
}7

1

Start of the probe.

2

Event begin (the start of the SystemTap session).

3

Start of the handler definition, indicated by {.

4

First function defined in the handler: the printf function.

5

String to be printed by the printf function, followed by a line break (/n).

6

Second function defined in the handler: the exit() function. Note that the SystemTap script will continue to run until the exit() function executes. If you want to stop the execution of the script before, stop it manually by pressing CtrlC.

7

End of the handler definition, indicated by }.

The event begin 2 (the start of the SystemTap session) triggers the handler enclosed in { }. Here, that is the printf function 4. In this case, it prints hello world followed by a new line 5. Then, the script exits.

If your statement block holds several statements, SystemTap executes these statements in sequence—you do not need to insert special separators or terminators between multiple statements. A statement block can also be nested within another statement blocks. Generally, statement blocks in SystemTap scripts use the same syntax and semantics as in the C programming language.

4.3.2 SystemTap Events (Probe Points)

SystemTap supports several built-in events.

The general event syntax is a dotted-symbol sequence. This allows a breakdown of the event namespace into parts. Each component identifier may be parametrized by a string or number literal, with a syntax like a function call. A component may include a * character, to expand to other matching probe points. A probe point may be followed by a ? character, to indicate that it is optional, and that no error should result if it fails to expand. Alternately, a probe point may be followed by a ! character to indicate that it is both optional and sufficient.

SystemTap supports multiple events per probe—they need to be separated by a comma (,). If multiple events are specified in a single probe, SystemTap will execute the handler when any of the specified events occur.

In general, events can be classified into the following categories:

  • Synchronous events: Occur when any process executes an instruction at a particular location in kernel code. This gives other events a reference point (instruction address) from which more contextual data may be available.

    An example for a synchronous event is vfs.FILE_OPERATION: The entry to the FILE_OPERATION event for Virtual File System (VFS). For example, in Section 4.2, “Installation and Setup”, read is the FILE_OPERATION event used for VFS.

  • Asynchronous events: Not tied to a particular instruction or location in code. This family of probe points consists mainly of counters, timers, and similar constructs.

    Examples for asynchronous events are: begin (start of a SystemTap session—when a SystemTap script is run, end (end of a SystemTap session), or timer events. Timer events specify a handler to be executed periodically, like example timer.s(SECONDS), or timer.ms(MILLISECONDS).

    When used together with other probes that collect information, timer events allow you to print periodic updates and see how that information changes over time.

Example 4.2: Probe with Timer Event

For example, the following probe would print the text hello world every 4 seconds:

probe timer.s(4)
{
   printf("hello world\n")
}

For detailed information about supported events, refer to the stapprobes man page. The See Also section of the man page also contains links to other man pages that discuss supported events for specific subsystems and components.

4.3.3 SystemTap Handlers (Probe Body)

Each SystemTap event is accompanied by a corresponding handler defined for that event, consisting of a statement block.

4.3.3.1 Functions

If you need the same set of statements in multiple probes, you can place them in a function for easy reuse. Functions are defined by the keyword function followed by a name. They take any number of string or numeric arguments (by value) and may return a single string or number.

function FUNCTION_NAME(ARGUMENTS) {STATEMENTS}
probe EVENT {FUNCTION_NAME(ARGUMENTS)}

The statements in FUNCTION_NAME are executed when the probe for EVENT executes. The ARGUMENTS are optional values passed into the function.

Functions can be defined anywhere in the script. They may take any

One of the functions needed very often was already introduced in Example 4.1, “Simple SystemTap Script”: the printf function for printing data in a formatted way. When using the printf function, you can specify how arguments should be printed by using a format string. The format string is included in quotation marks and can contain further format specifiers, introduced by a % character.

Which format strings to use depends on your list of arguments. Format strings can have multiple format specifiers—each matching a corresponding argument. Multiple arguments can be separated by a comma.

Example 4.3: printf Function with Format Specifiers
printf ("1%s2(%d3) open\n4", execname(), pid())

1

Start of the format string, indicated by ".

2

String format specifier.

3

Integer format specifier.

4

End of the format string, indicated by ".

The example above prints the current executable name (execname()) as a string and the process ID (pid()) as an integer in brackets. Then, a space, the word open and a line break follow:

[...]
vmware-guestd(2206) open
hald(2360) open
[...]

Apart from the two functions execname()and pid()) used in Example 4.3, “printf Function with Format Specifiers”, a variety of other functions can be used as printf arguments.

Among the most commonly used SystemTap functions are the following:

tid()

ID of the current thread.

pid()

Process ID of the current thread.

uid()

ID of the current user.

cpu()

Current CPU number.

execname()

Name of the current process.

gettimeofday_s()

Number of seconds since Unix epoch (January 1, 1970).

ctime()

Convert time into a string.

pp()

String describing the probe point currently being handled.

thread_indent()

Useful function for organizing print results. It (internally) stores an indentation counter for each thread (tid()). The function takes one argument, an indentation delta, indicating how many spaces to add or remove from the thread's indentation counter. It returns a string with some generic trace data along with an appropriate number of indentation spaces. The generic data returned includes a time stamp (number of microseconds since the initial indentation for the thread), a process name, and the thread ID itself. This allows you to identify what functions were called, who called them, and how long they took.

Call entries and exits often do not immediately precede each other (otherwise it would be easy to match them). In between a first call entry and its exit, usually other call entries and exits are made. The indentation counter helps you match an entry with its corresponding exit as it indents the next function call in case it is not the exit of the previous one. For an example SystemTap script using thread_indent() and the respective output, refer to the SystemTap Tutorial: https://sourceware.org/systemtap/tutorial.pdf.

For more information about supported SystemTap functions, refer to the stapfuncs man page.

4.3.3.2 Other Basic Constructs

Apart from functions, you can use other common constructs in SystemTap handlers, including variables, conditional statements (like if/else, while loops, for loops, arrays or command line arguments.

4.3.3.2.1 Variables

Variables may be defined anywhere in the script. To define one, simply choose a name and assign a value from a function or expression to it:

foo = gettimeofday( )

Then you can use the variable in an expression. From the type of values assigned to the variable, SystemTap automatically infers the type of each identifier (string or number). Any inconsistencies will be reported as errors. In the example above, foo would automatically be classified as a number and could be printed via printf() with the integer format specifier (%d).

However, by default, variables are local to the probe they are used in: They are initialized, used and disposed of at each handler evocation. To share variables between probes, declare them global anywhere in the script. To do so, use the global keyword outside of the probes:

Example 4.4: Using Global Variables
global count_jiffies, count_ms
probe timer.jiffies(100) { count_jiffies ++ }
probe timer.ms(100) { count_ms ++ }
probe timer.ms(12345)
{
  hz=(1000*count_jiffies) / count_ms
  printf ("jiffies:ms ratio %d:%d => CONFIG_HZ=%d\n",
    count_jiffies, count_ms, hz)
  exit ()
  }

This example script computes the CONFIG_HZ setting of the kernel by using timers that count jiffies and milliseconds, then computing accordingly. (A jiffy is the duration of one tick of the system timer interrupt. It is not an absolute time interval unit, since its duration depends on the clock interrupt frequency of the particular hardware platform). With the global statement it is possible to use the variables count_jiffies and count_ms also in the probe timer.ms(12345). With ++ the value of a variable is incremented by 1.

4.3.3.2.2 Conditional Statements

There are several conditional statements that you can use in SystemTap scripts. The following are probably the most common:

If/Else Statements

They are expressed in the following format:

if (CONDITION)1STATEMENT12
else3STATEMENT24

The if statement compares an integer-valued expression to zero. If the condition expression 1 is non-zero, the first statement 2 is executed. If the condition expression is zero, the second statement 4 is executed. The else clause (3 and 4) is optional. Both 2 and 4 can also be statement blocks.

While Loops

They are expressed in the following format:

while (CONDITION)1STATEMENT2

As long as condition is non-zero, the statement 2 is executed. 2 can also be a statement block. It must change a value so condition will eventually be zero.

For Loops

They are a shortcut for while loops and are expressed in the following format:

for (INITIALIZATION1; CONDITIONAL2; INCREMENT3) statement

The expression specified in 1 is used to initialize a counter for the number of loop iterations and is executed before execution of the loop starts. The execution of the loop continues until the loop condition 2 is false. (This expression is checked at the beginning of each loop iteration). The expression specified in 3 is used to increment the loop counter. It is executed at the end of each loop iteration.

Conditional Operators

The following operators can be used in conditional statements:

==: Is equal to

!=: Is not equal to

>=: Is greater than or equal to

<=: Is less than or equal to

4.4 Example Script

If you have installed the systemtap-docs package, you can find several useful SystemTap example scripts in /usr/share/doc/packages/systemtap/examples.

This section describes a rather simple example script in more detail: /usr/share/doc/packages/systemtap/examples/network/tcp_connections.stp.

Example 4.5: Monitoring Incoming TCP Connections with tcp_connections.stp
#! /usr/bin/env stap

probe begin {
  printf("%6s %16s %6s %6s %16s\n",
         "UID", "CMD", "PID", "PORT", "IP_SOURCE")
}

probe kernel.function("tcp_accept").return?,
      kernel.function("inet_csk_accept").return? {
  sock = $return
  if (sock != 0)
    printf("%6d %16s %6d %6d %16s\n", uid(), execname(), pid(),
           inet_get_local_port(sock), inet_get_ip_source(sock))
}

This SystemTap script monitors the incoming TCP connections and helps to identify unauthorized or unwanted network access requests in real time. It shows the following information for each new incoming TCP connection accepted by the computer:

  • User ID (UID)

  • Command accepting the connection (CMD)

  • Process ID of the command (PID)

  • Port used by the connection (PORT)

  • IP address from which the TCP connection originated (IP_SOUCE)

To run the script, execute

stap /usr/share/doc/packages/systemtap/examples/network/tcp_connections.stp

and follow the output on the screen. To manually stop the script, press CtrlC.

4.5 User Space Probing

For debugging user space applications (like DTrace can do), SUSE Linux Enterprise Server 12 SP5 supports user space probing with SystemTap: Custom probe points can be inserted in any user space application. Thus, SystemTap lets you use both kernel space and user space probes to debug the behavior of the whole system.

To get the required utrace infrastructure and the uprobes kernel module for user space probing, you need to install the kernel-trace package in addition to the packages listed in Section 4.2, “Installation and Setup”.

utrace implements a framework for controlling user space tasks. It provides an interface that can be used by various tracing engines, implemented as loadable kernel modules. The engines register callback functions for specific events, then attach to whichever thread they want to trace. As the callbacks are made from safe places in the kernel, this allows for great leeway in the kinds of processing the functions can do. Various events can be watched via utrace, for example, system call entry and exit, fork(), signals being sent to the task, etc. More details about the utrace infrastructure are available at https://sourceware.org/systemtap/wiki/utrace.

SystemTap includes support for probing the entry into and return from a function in user space processes, probing predefined markers in user space code, and monitoring user-process events.

To check if the currently running kernel provides the needed utrace support, use the following command:

 grep CONFIG_UTRACE /boot/config-`uname -r`

For more details about user space probing, refer to https://sourceware.org/systemtap/SystemTap_Beginners_Guide/userspace-probing.html.

4.6 For More Information

This chapter only provides a short SystemTap overview. Refer to the following links for more information about SystemTap:

https://sourceware.org/systemtap/

SystemTap project home page.

https://sourceware.org/systemtap/wiki/

Huge collection of useful information about SystemTap, ranging from detailed user and developer documentation to reviews and comparisons with other tools, or Frequently Asked Questions and tips. Also contains collections of SystemTap scripts, examples and usage stories and lists recent talks and papers about SystemTap.

https://sourceware.org/systemtap/documentation.html

Features a SystemTap Tutorial, a SystemTap Beginner's Guide, a Tapset Developer's Guide, and a SystemTap Language Reference in PDF and HTML format. Also lists the relevant man pages.

You can also find the SystemTap language reference and SystemTap tutorial in your installed system under /usr/share/doc/packages/systemtap. Example SystemTap scripts are available from the example subdirectory.

5 Kernel Probes

Kernel probes are a set of tools to collect Linux kernel debugging and performance information. Developers and system administrators usually use them either to debug the kernel, or to find system performance bottlenecks. The reported data can then be used to tune the system for better performance.

You can insert these probes into any kernel routine, and specify a handler to be invoked after a particular break-point is hit. The main advantage of kernel probes is that you no longer need to rebuild the kernel and reboot the system after you make changes in a probe.

To use kernel probes, you typically need to write or obtain a specific kernel module. Such modules include both the init and the exit function. The init function (such as register_kprobe()) registers one or more probes, while the exit function unregisters them. The registration function defines where the probe will be inserted and which handler will be called after the probe is hit. To register or unregister a group of probes at one time, you can use relevant register_<PROBE_TYPE>probes() or unregister_<PROBE_TYPE>probes() functions.

Debugging and status messages are typically reported with the printk kernel routine. printk is a kernel space equivalent of a user space printf routine. For more information on printk, see Logging kernel messages. Normally, you can view these messages by inspecting the output of the systemd journal (see Book “Administration Guide”, Chapter 16 “journalctl: Query the systemd Journal”). For more information on log files, see Chapter 3, Analyzing and Managing System Log Files.

5.1 Supported Architectures

Kernel probes are fully implemented on the following architectures:

  • x86

  • AMD64/Intel 64

  • Arm

  • POWER

Kernel probes are partially implemented on the following architectures:

  • IA64 (does not support probes on instruction slot1)

  • sparc64 (return probes not yet implemented)

5.2 Types of Kernel Probes

There are three types of kernel probes: Kprobes, Jprobes, and Kretprobes. Kretprobes are sometimes called return probes. You can find source code examples of all three type of probes in the Linux kernel. See the directory /usr/src/linux/samples/kprobes/ (package kernel-source).

5.2.1 Kprobes

Kprobes can be attached to any instruction in the Linux kernel. When Kprobes is registered, it inserts a break-point at the first byte of the probed instruction. When the processor hits this break-point, the processor registers are saved, and the processing passes to Kprobes. First, a pre-handler is executed, then the probed instruction is stepped, and, finally a post-handler is executed. The control is then passed to the instruction following the probe point.

5.2.2 Jprobes

Jprobes is implemented through the Kprobes mechanism. It is inserted on a function's entry point and allows direct access to the arguments of the function which is being probed. Its handler routine must have the same argument list and return value as the probed function. To end it, call the jprobe_return() function.

When a jprobe is hit, the processor registers are saved, and the instruction pointer is directed to the jprobe handler routine. The control then passes to the handler with the same register contents as the function being probed. Finally, the handler calls the jprobe_return() function, and switches the control back to the control function.

In general, you can insert multiple probes on one function. Jprobe is, however, limited to only one instance per function.

5.2.3 Return Probe

Return probes are also implemented through Kprobes. When the register_kretprobe() function is called, a kprobe is attached to the entry of the probed function. After hitting the probe, the kernel probes mechanism saves the probed function return address and calls a user-defined return handler. The control is then passed back to the probed function.

Before you call register_kretprobe(), you need to set a maxactive argument, which specifies how many instances of the function can be probed at the same time. If set too low, you will miss a certain number of probes.

5.3 Kprobes API

The programming interface of Kprobes consists of functions which are used to register and unregister all used kernel probes, and associated probe handlers. For a more detailed description of these functions and their arguments, see the information sources in Section 5.5, “For More Information”.

register_kprobe()

Inserts a break-point on a specified address. When the break-point is hit, the pre_handler and post_handler are called.

register_jprobe()

Inserts a break-point in the specified address. The address needs to be the address of the first instruction of the probed function. When the break-point is hit, the specified handler is run. The handler should have the same argument list and return type as the probed.

register_kretprobe()

Inserts a return probe for the specified function. When the probed function returns, a specified handler is run. This function returns 0 on success, or a negative error number on failure.

unregister_kprobe(), unregister_jprobe(), unregister_kretprobe()

Removes the specified probe. You can use it any time after the probe has been registered.

register_kprobes(), register_jprobes(), register_kretprobes()

Inserts each of the probes in the specified array.

unregister_kprobes(), unregister_jprobes(), unregister_kretprobes()

Removes each of the probes in the specified array.

disable_kprobe(), disable_jprobe(), disable_kretprobe()

Disables the specified probe temporarily.

enable_kprobe(), enable_jprobe(), enable_kretprobe()

Temporarily enables disabled probes.

5.4 debugfs Interface

In recent Linux kernels, the Kprobes instrumentation uses the kernel's debugfs interface. It can list all registered probes and globally switch all probes on or off.

5.4.1 Listing Registered Kernel Probes

The list of all currently registered probes is in the /sys/kernel/debug/kprobes/list file.

saturn.example.com:~ # cat /sys/kernel/debug/kprobes/list
c015d71a  k  vfs_read+0x0   [DISABLED]
c011a316  j  do_fork+0x0
c03dedc5  r  tcp_v4_rcv+0x0

The first column lists the address in the kernel where the probe is inserted. The second column prints the type of the probe: k for kprobe, j for jprobe, and r for return probe. The third column specifies the symbol, offset and optional module name of the probe. The following optional columns include the status information of the probe. If the probe is inserted on a virtual address which is not valid anymore, it is marked with [GONE]. If the probe is temporarily disabled, it is marked with [DISABLED].

5.4.2 How to Switch All Kernel Probes On or Off

The /sys/kernel/debug/kprobes/enabled file represents a switch with which you can globally and forcibly turn on or off all the registered kernel probes. To turn them off, simply enter

echo "0" > /sys/kernel/debug/kprobes/enabled

on the command line as root. To turn them on again, enter

echo "1" > /sys/kernel/debug/kprobes/enabled

Note that this way you do not change the status of the probes. If a probe is temporarily disabled, it will not be enabled automatically but will remain in the [DISABLED] state after entering the latter command.

5.5 For More Information

To learn more about kernel probes, look at the following sources of information:

  • Thorough but more technically oriented information about kernel probes is in /usr/src/linux/Documentation/kprobes.txt (package kernel-source).

  • Examples of all three types of probes (together with related Makefile) are in the /usr/src/linux/samples/kprobes/ directory (package kernel-source).

  • In-depth information about Linux kernel modules and printk kernel routine can be found at The Linux Kernel Module Programming Guide

  • Practical but slightly outdated information about the use of kernel probes can be found in Kernel debugging with Kprobes

6 Hardware-Based Performance Monitoring with Perf

Perf is an interface to access the performance monitoring unit (PMU) of a processor and to record and display software events such as page faults. It supports system-wide, per-thread, and KVM virtualization guest monitoring.

You can store resulting information in a report. This report contains information about, for example, instruction pointers or what code a thread was executing.

Perf consists of two parts:

  • Code integrated into the Linux kernel that is responsible for instructing the hardware.

  • The perf user space utility that allows you to use the kernel code and helps you analyze gathered data.

6.1 Hardware-Based Monitoring

Performance monitoring means collecting information related to how an application or system performs. This information can be obtained either through software-based means or from the CPU or chipset. Perf integrates both of these methods.

Many modern processors contain a performance monitoring unit (PMU). The design and functionality of a PMU is CPU-specific. For example, the number of registers, counters and features supported will vary by CPU implementation.

Each PMU model consists of a set of registers: the performance monitor configuration (PMC) and the performance monitor data (PMD). Both can be read, but only PMCs are writable. These registers store configuration information and data.

6.2 Sampling and Counting

Perf supports several profiling modes:

  • Counting.  Count the number of occurrences of an event.

  • Event-Based Sampling.  A less exact way of counting: A sample is recorded whenever a certain threshold number of events has occurred.

  • Time-Based Sampling.  A less exact way of counting: A sample is recorded in a defined frequency.

  • Instruction-Based Sampling (AMD64 only).  The processor follows instructions appearing in a given interval and samples which events they produce. This allows following up on individual instructions and seeing which of them is critical to performance.

6.3 Installing Perf

The Perf kernel code is already included with the default kernel. To be able to use the user space utility, install the package perf.

6.4 Perf Subcommands

To gather the required information, the perf tool has several subcommands. This section gives an overview of the most often used commands.

To see help in the form of a man page for any of the subcommands, use either perf helpSUBCOMMAND or man perf-SUBCOMMAND.

perf stat

Start a program and create a statistical overview that is displayed after the program quits. perf stat is used to count events.

perf record

Start a program and create a report with performance counter information. The report is stored as perf.data in the current directory. perf record is used to sample events.

perf report

Display a report that was previously created with perf record.

perf annotate

Display a report file and an annotated version of the executed code. If debug symbols are installed, you will also see the source code displayed.

perf list

List event types that Perf can report with the current kernel and with your CPU. You can filter event types by category—for example, to see hardware events only, use perf list hw.

The man page for perf_event_open has short descriptions for the most important events. For example, to find a description of the event branch-misses, search for BRANCH_MISSES (note the spelling differences):

tux > man perf_event_open | grep -A5 BRANCH_MISSES

Sometimes, events may be ambiguous. Note that the lowercase hardware event names are not the name of raw hardware events but instead the name of aliases created by Perf. These aliases map to differently named but similarly defined hardware events on each supported processor.

For example, the cpu-cycles event is mapped to the hardware event UNHALTED_CORE_CYCLES on Intel processors. On AMD processors, however, it is mapped to the hardware event CPU_CLK_UNHALTED.

Perf also allows measuring raw events specific to your hardware. To look up their descriptions, see the Architecture Software Developer's Manual of your CPU vendor. The relevant documents for AMD64/Intel 64 processors are linked to in Section 6.7, “For More Information”.

perf top

Display system activity as it happens.

perf trace

This command behaves similarly to strace. With this subcommand, you can see which system calls are executed by a particular thread or process and which signals it receives.

6.5 Counting Particular Types of Event

To count the number of occurrences of an event, such as those displayed by perf list, use:

root # perf stat -e EVENT -a

To count multiple types of events at once, list them separated by commas. For example, to count cpu-cycles and instructions, use:

root # perf stat -e cpu-cycles,instructions -a

To stop the session, press CtrlC.

You can also count the number of occurrences of an event within a particular time:

root # perf stat -e EVENT -a -- sleep TIME

Replace TIME by a value in seconds.

6.6 Recording Events Specific to Particular Commands

There are various ways to sample events specific to a particular command:

  • To create a report for a newly invoked command, use:

    root # perf record COMMAND

    Then, use the started process normally. When you quit the process, the Perf session will also stop.

  • To create a report for the entire system while a newly invoked command is running, use:

    root # perf record -a COMMAND

    Then, use the started process normally. When you quit the process, the Perf session will also stop.

  • To create a report for an already running process, use:

    root # perf record -p PID

    Replace PID with a process ID. To stop the session, press CtrlC.

Now you can view the gathered data (perf.data) using:

tux > perf report

This will open a pseudo-graphical interface. To receive help, press H. To quit, press Q.

If you prefer a graphical interface, try the GTK+ interface of Perf:

tux > perf report --gtk

However, note that the GTK+ interface is very limited in functionality.

6.7 For More Information

This chapter only provides a short overview. Refer to the following links for more information:

https://perf.wiki.kernel.org/index.php/Main_Page

The project home page. It also features a tutorial on using perf.

http://www.brendangregg.com/perf.html

Unofficial page with many one-line examples of how to use perf.

http://web.eece.maine.edu/~vweaver/projects/perf_events/

Unofficial page with several resources, mostly relating to the Linux kernel code of Perf and its API. This page includes, for example, a CPU compatibility table and a programming guide.

https://www-ssl.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-vol-3b-part-2-manual.pdf

The Intel Architectures Software Developer's Manual, Volume 3B.

https://support.amd.com/TechDocs/24593.pdf

The AMD Architecture Programmer's Manual, Volume 2.

Chapter 7, OProfile—System-Wide Profiler

Consult this chapter for other performance optimizations.

7 OProfile—System-Wide Profiler

OProfile is a profiler for dynamic program analysis. It investigates the behavior of a running program and gathers information. This information can be viewed and gives hints for further optimization.

It is not necessary to recompile or use wrapper libraries to use OProfile. Not even a kernel patch is needed. Usually, when profiling an application, a small overhead is expected, depending on the workload and sampling frequency.

7.1 Conceptual Overview

OProfile consists of a kernel driver and a daemon for collecting data. It uses the hardware performance counters provided on many processors. OProfile is capable of profiling all code including the kernel, kernel modules, kernel interrupt handlers, system shared libraries, and other applications.

Modern processors support profiling through the hardware by performance counters. Depending on the processor, there can be many counters and each of these can be programmed with an event to count. Each counter has a value which determines how often a sample is taken. The lower the value, the more often it is used.

During the post-processing step, all information is collected and instruction addresses are mapped to a function name.

7.2 Installation and Requirements

To use OProfile, install the oprofile package that is included with the SLE SDK. OProfile works on AMD64/Intel 64, IBM IBM Z, and IBM POWER processors. To find out how to install software from the SDK, refer to Book “Deployment Guide”, Chapter 15 “Installing Modules, Extensions, and Third Party Add-On Products”, Section 15.5 “SUSE Software Development Kit (SDK) 12 SP5.

It is useful to install the *-debuginfo package for the respective application you want to profile. If you want to profile the kernel, you need the debuginfo package as well.

7.3 Available OProfile Utilities

OProfile contains several utilities to handle the profiling process and its profiled data. The following list is a short summary of programs used in this chapter:

opannotate

Outputs annotated source or assembly listings mixed with profile information. An annotated report can be used in combination with addr2line to identify the source file and line where hotspots potentially exist. See man addr2line for more information.

opcontrol

Controls the profiling sessions (start or stop), dumps profile data, and sets up parameters.

ophelp

Lists available events with short descriptions.

opimport

Converts sample database files from a foreign binary format to the native format.

opreport

Generates reports from profiled data.

7.4 Using OProfile

With OProfile, you can profile both the kernel and applications. When profiling the kernel, tell OProfile where to find the vmlinuz* file. Use the --vmlinux option and point it to vmlinuz* (usually in /boot). If you need to profile kernel modules, OProfile does this by default. However, make sure you read http://oprofile.sourceforge.net/doc/kernel-profiling.html.

Applications usually do not need to profile the kernel, therefore you should use the --no-vmlinux option to reduce the amount of information.

7.4.1 Creating a Report

Starting the daemon, collecting data, stopping the daemon, and creating a report.

  1. Open a shell and log in as root.

  2. Decide if you want to profile with or without the Linux kernel:

    1. Profile With the Linux Kernel.  Execute the following commands, because opcontrol can only work with uncompressed images:

      cp /boot/vmlinux-`uname -r`.gz /tmp
      gunzip /tmp/vmlinux*.gz
      opcontrol --vmlinux=/tmp/vmlinux*
    2. Profile Without the Linux Kernel.  Use the following command:

      opcontrol --no-vmlinux

      To see which functions call other functions in the output, additionally use the --callgraph option and set a maximum DEPTH:

      opcontrol --no-vmlinux --callgraph DEPTH
  3. Start the OProfile daemon:

    opcontrol --start
    Using 2.6+ OProfile kernel interface.
    Using log file /var/lib/oprofile/samples/oprofiled.log
    Daemon started.
    Profiler running.
  4. Now start the application you want to profile.

  5. Stop the OProfile daemon:

    opcontrol --stop
  6. Dump the collected data to /var/lib/oprofile/samples:

    opcontrol --dump
  7. Create a report:

    opreport
    Overflow stats not available
    CPU: CPU with timer interrupt, speed 0 MHz (estimated)
    Profiling through timer interrupt
              TIMER:0|
      samples|      %|
    ------------------
        84877 98.3226 no-vmlinux
    ...
  8. Shut down the oprofile daemon:

    opcontrol --shutdown

7.4.2 Getting Event Configurations

The general procedure for event configuration is as follows:

  1. Use first the events CPU-CLK_UNHALTED and INST_RETIRED to find optimization opportunities.

  2. Use specific events to find bottlenecks. To list them, use the command opcontrol --list-events.

If you need to profile certain events, first check the available events supported by your processor with the ophelp command (example output generated from Intel Core i5 CPU):

ophelp
oprofile: available events for CPU type "Intel Architectural Perfmon"

See Intel 64 and IA-32 Architectures Software Developer's Manual
Volume 3B (Document 253669) Chapter 18 for architectural perfmon events
This is a limited set of fallback events because oprofile does not know your CPU
CPU_CLK_UNHALTED: (counter: all))
        Clock cycles when not halted (min count: 6000)
INST_RETIRED: (counter: all))
        number of instructions retired (min count: 6000)
LLC_MISSES: (counter: all))
        Last level cache demand requests from this core that missed the LLC (min count: 6000)
        Unit masks (default 0x41)
        ----------
        0x41: No unit mask
LLC_REFS: (counter: all))
        Last level cache demand requests from this core (min count: 6000)
        Unit masks (default 0x4f)
        ----------
        0x4f: No unit mask
BR_MISS_PRED_RETIRED: (counter: all))
        number of mispredicted branches retired (precise) (min count: 500)

You can get the same output from opcontrol --list-events.

Specify the performance counter events with the option --event. Multiple options are possible. This option needs an event name (from ophelp) and a sample rate, for example:

opcontrol --event=CPU_CLK_UNHALTED:100000
Warning
Warning: Setting Sampling Rates with CPU_CLK_UNHALTED

Setting low sampling rates can seriously impair the system performance while high sample rates can disrupt the system to such a high degree that the data is useless. It is recommended to tune the performance metric for being monitored with and without OProfile and to experimentally determine the minimum sample rate that disrupts the performance the least.

7.5 Using OProfile's GUI

The GUI for OProfile can be started as root with oprof_start, see Figure 7.1, “GUI for OProfile”. Select your events and change the counter, if necessary. Every green line is added to the list of checked events. Hover the mouse over the line to see a help text in the status line below. Use the Configuration tab to set the buffer and CPU size, the verbose option and others. Click Start to execute OProfile.

GUI for OProfile
Figure 7.1: GUI for OProfile

7.6 Generating Reports

Before generating a report, make sure OProfile has dumped your data to the /var/lib/oprofile/samples directory using the command opcontrol --dump. A report can be generated with the commands opreport or opannotate.

Calling opreport without any options gives a complete summary. With an executable as an argument, retrieve profile data only from this executable. If you analyze applications written in C++, use the --demangle smart option.

The opannotate generates output with annotations from source code. Run it with the following options:

opannotate --source \
   --base-dirs=BASEDIR \
   --search-dirs=SEARCHDIR \
   --output-dir=annotated/ \
   /lib/libfoo.so

The option --base-dir contains a comma separated list of paths which is stripped from debug source files. These paths were searched prior to looking in --search-dirs. The --search-dirs option is also a comma separated list of directories to search for source files.

Note
Note: Inaccuracies in Annotated Source

Because of compiler optimization, code can disappear and appear in a different place. Use the information in http://oprofile.sourceforge.net/doc/debug-info.html to fully understand its implications.

7.7 For More Information

This chapter only provides a short overview. Refer to the following links for more information:

http://oprofile.sourceforge.net

The project home page.

Manpages

Details descriptions about the options of the different tools.

/usr/share/doc/packages/oprofile/oprofile.html

Contains the OProfile manual.

http://developer.intel.com/

Architecture reference for Intel processors.

Part IV Resource Management

  • 8 General System Resource Management
  • Tuning the system is not only about optimizing the kernel or getting the most out of your application, it begins with setting up a lean and fast system. The way you set up your partitions and file systems can influence the server's speed. The number of active services and the way routine tasks are scheduled also affects performance.

  • 9 Kernel Control Groups
  • Kernel Control Groups (cgroups) are a kernel feature that allows assigning and limiting hardware and system resources for processes. Processes can also be organized in a hierarchical tree structure.

  • 10 Automatic Non-Uniform Memory Access (NUMA) Balancing
  • There are physical limitations to hardware that are encountered when many CPUs and lots of memory are required. In this chapter, the important limitation is that there is limited communication bandwidth between the CPUs and the memory. One architecture modification that was introduced to address this is Non-Uniform Memory Access (NUMA).

    In this configuration, there are multiple nodes. Each of the nodes contains a subset of all CPUs and memory. The access speed to main memory is determined by the location of the memory relative to the CPU. The performance of a workload depends on the application threads accessing data that is local to the CPU the thread is executing on. Automatic NUMA Balancing is a new feature of SLE 12. Automatic NUMA Balancing migrates data on demand to memory nodes that are local to the CPU accessing that data. Depending on the workload, this can dramatically boost performance when using NUMA hardware.

  • 11 Power Management
  • Power management aims at reducing operating costs for energy and cooling systems while at the same time keeping the performance of a system at a level that matches the current requirements. Thus, power management is always a matter of balancing the actual performance needs and power saving options for a system. Power management can be implemented and used at different levels of the system. A set of specifications for power management functions of devices and the operating system interface to them has been defined in the Advanced Configuration and Power Interface (ACPI). As power savings in server environments can primarily be achieved at the processor level, this chapter introduces some main concepts and highlights some tools for analyzing and influencing relevant parameters.

8 General System Resource Management

Tuning the system is not only about optimizing the kernel or getting the most out of your application, it begins with setting up a lean and fast system. The way you set up your partitions and file systems can influence the server's speed. The number of active services and the way routine tasks are scheduled also affects performance.

8.1 Planning the Installation

A carefully planned installation ensures that the system is set up exactly as you need it for the given purpose. It also saves considerable time when fine tuning the system. All changes suggested in this section can be made in the Installation Settings step during the installation. See Book “Deployment Guide”, Chapter 6 “Installation with YaST”, Section 6.15 “Installation Settings” for details.

8.1.1 Partitioning

Depending on the server's range of applications and the hardware layout, the partitioning scheme can influence the machine's performance (although to a lesser extent only). It is beyond the scope of this manual to suggest different partitioning schemes for particular workloads. However, the following rules will positively affect performance. They do not apply when using an external storage system.

  • Make sure there always is some free space available on the disk, since a full disk delivers inferior performance

  • Disperse simultaneous read and write access onto different disks by, for example:

    • using separate disks for the operating system, data, and log files

    • placing a mail server's spool directory on a separate disk

    • distributing the user directories of a home server between different disks

8.1.2 Installation Scope

The installation scope has no direct influence on the machine's performance, but a carefully chosen scope of packages has advantages. It is recommended to install the minimum of packages needed to run the server. A system with a minimum set of packages is easier to maintain and has fewer potential security issues. Furthermore, a tailor made installation scope also ensures that no unnecessary services are started by default.

SUSE Linux Enterprise Server lets you customize the installation scope on the Installation Summary screen. By default, you can select or remove preconfigured patterns for specific tasks, but it is also possible to start the YaST Software Manager for a fine-grained package-based selection.

One or more of the following default patterns may not be needed in all cases:

GNOME Desktop Environment

Servers rarely need a full desktop environment. In case a graphical environment is needed, a more economical solution such as IceWM can be sufficient.

X Window System

When solely administrating the server and its applications via command line, consider not installing this pattern. However, keep in mind that it is needed to run GUI applications from a remote machine. If your application is managed by a GUI or if you prefer the GUI version of YaST, keep this pattern.

Print Server

This pattern is only needed if you want to print from the machine.

8.1.3 Default Target

A running X Window System consumes many resources and is rarely needed on a server. It is strongly recommended to start the system in target multi-user.target. You will still be able to remotely start graphical applications.

8.2 Disabling Unnecessary Services

The default installation starts several services (the number varies with the installation scope). Since each service consumes resources, it is recommended to disable the ones not needed. Run YaST › System › Services Manager to start the services management module.

If you are using the graphical version of YaST, you can click the column headlines to sort the list of services. Use this to get an overview of which services are currently running. Use the Start/Stop button to disable the service for the running session. To permanently disable it, use the Enable/Disable button.

The following list shows services that are started by default after the installation of SUSE Linux Enterprise Server. Check which of the components you need, and disable the others:

alsasound

Loads the Advanced Linux Sound System.

auditd

A daemon for the Audit system (see Book “Security and Hardening Guide” for details). Disable this if you do not use Audit.

bluez-coldplug

Handles cold plugging of Bluetooth dongles.

cups

A printer daemon.

java.binfmt_misc

Enables the execution of *.class or *.jar Java programs.

nfs

Services needed to mount NFS.

smbfs

Services needed to mount SMB/CIFS file systems from a Windows* server.

splash / splash_early

Shows the splash screen on start-up.

8.3 File Systems and Disk Access

Hard disks are the slowest components in a computer system and therefore often the cause for a bottleneck. Using the file system that best suits your workload helps to improve performance. Using special mount options or prioritizing a process's I/O priority are further means to speed up the system.

8.3.1 File Systems

SUSE Linux Enterprise Server ships with several file systems, including Btrfs, Ext4, Ext3, Ext2, ReiserFS, and XFS. Each file system has its own advantages and disadvantages. Refer to Book “Storage Administration Guide”, Chapter 1 “Overview of File Systems in Linux” for detailed information.

8.3.1.1 NFS

NFS (Version 3) tuning is covered in detail in the NFS Howto at http://nfs.sourceforge.net/nfs-howto/. The first thing to experiment with when mounting NFS shares is increasing the read write blocksize to 32768 by using the mount options wsize and rsize.

8.3.2 Time Stamp Update Policy

Each file and directory in a file system has three time stamps associated with it: a time when the file was last read called access time, a time when the file data was last modified called modification time, and a time when the file metadata was last modified called change time. Keeping access time always up to date has significant performance overhead since every read-only access will incur a write operation. Thus by default every file system updates access time only if current file access time is older than a day or if it is older than file modification or change time. This feature is called relative access time and the corresponding mount option is relatime. Updates of access time can be completely disabled using the noatime mount option, however you need to verify your applications do not use it. This can be true for file and Web servers or for network storage. If the default relative access time update policy is not suitable for your applications, use the strictatime mount option.

Some file systems (for example Ext4) also support lazy time stamp updates. When this feature is enabled using the lazytime mount option, updates of all time stamps happen in memory but they are not written to disk. That happens only in response to fsync or sync system calls, when the file information is written due to another reason such as file size update, when time stamps are older than 24 hours, or when cached file information needs to be evicted from memory.

To update mount options used for a file system, either edit /etc/fstab directly, or use the Fstab Options dialog when editing or adding a partition with the YaST Partitioner.

8.3.3 Prioritizing Disk Access with ionice

The ionice command lets you prioritize disk access for single processes. This enables you to give less I/O priority to background processes with heavy disk access that are not time-critical, such as backup jobs. ionice also lets you raise the I/O priority for a specific process to make sure this process always has immediate access to the disk. The caveat with this feature is that standard writes are cached in the page cache and are written back to persistent storage only later by an independent kernel process. Thus, the I/O priority setting generally does not apply for these writes. Also be aware that the I/O class and priority settings are obeyed only by the CFQ I/O scheduler for the legacy block I/O path (refer to Section 12.2, “Available I/O Elevators”), and the BFQ I/O scheduler for the blk-mq I/O path (refer to Section 12.3, “Available I/O Elevators with blk-mq I/O path”). You can set the following three scheduling classes:

Idle

A process from the idle scheduling class is only granted disk access when no other process has asked for disk I/O.

Best effort

The default scheduling class used for any process that has not asked for a specific I/O priority. Priority within this class can be adjusted to a level from 0 to 7 (with 0 being the highest priority). Programs running at the same best-effort priority are served in a round-robin fashion. Some kernel versions treat priority within the best-effort class differently—for details, refer to the ionice(1) man page.

Real-time

Processes in this class are always granted disk access first. Fine-tune the priority level from 0 to 7 (with 0 being the highest priority). Use with care, since it can starve other processes.

For more details and the exact command syntax refer to the ionice(1) man page. If you need more reliable control over bandwidth available to each application, use Kernel Control Groups as described in Section 9.2, “Setting Resource Limits”.

9 Kernel Control Groups

Kernel Control Groups (cgroups) are a kernel feature that allows assigning and limiting hardware and system resources for processes. Processes can also be organized in a hierarchical tree structure.

9.1 Overview

Every process is assigned exactly one administrative cgroup. cgroups are ordered in a hierarchical tree structure. Resource limitations can be set for single processes or for whole branches of the hierarchy tree. Limitations for CPU, memory, disk I/O, or network bandwidth usage can be set.

On SUSE Linux Enterprise Server, systemd uses cgroups to organize all processes in groups, which systemd calls slices. systemd also provides an interface for setting cgroup properties.

The command systemd-cgls displays the hierarchy tree.

This chapter is an overview. For more details, refer to the listed references.

9.2 Setting Resource Limits

Note
Note: Implicit Resource Consumption

Be aware that resource consumption implicitly depends on the environment where your workload executes (e.g. size of data structures in libraries/kernel, forking behavior of utilities, computational efficiency), hence it is recommended to (re)calibrate your limits should the environment change.

Limitations for cgroups can be set with the systemctl set-property command. The syntax is:

root # systemctl set-property [--runtime] NAME PROPERTY1=VALUE [PROPERTY2=VALUE]

Optionally, use the --runtime option. With this option, set limits do not persist after the next reboot.

Replace NAME with a systemd service slice, scope, socket, mount, or swap name. Replace properties with one or more of the following:

CPUAccounting=[yes|no]

Turns on CPU usage accounting. This property takes yes and no as arguments.

Example:

root # systemctl set-property user.slice CPUAccounting=yes
CPUQuota=PERCENTAGE

Assigns a CPU time to processes. The value is a percentage followed by a % as suffix. This requires CPUAccounting=yes.

Example:

root # systemctl set-property user.slice CPUQuota=50%
MemoryAccounting=[yes|no]

Turns on memory usage accounting. This property takes yes and no as arguments.

Example:

root # systemctl set-property user.slice MemoryAccounting=yes
MemoryLow=BYTES

Unused memory from processes below this limit will not be reclaimed for other use. Use suffixes K, M, G or T for BYTES. This requires MemoryAccounting=yes.

Example:

root # systemctl set-property nginx.service MemoryLow=512M
MemoryHigh=BYTES

If more memory above this limit is used, memory is aggressively taken away from the processes. Use suffixes K, M, G or T for BYTES. This requires MemoryAccounting=yes.

Example:

root # systemctl set-property nginx.service MemoryHigh=2G
MemoryMax=BYTES

Sets a maximum limit for used memory. Processes will be killed if they use more memory than allowed. Use suffixes K, M, G or T for BYTES. This requires MemoryAccounting=yes.

Example:

root # systemctl set-property nginx.service MemoryMax=4G
DeviceAllow=

Allows read (r), write (w) and mknod (m) access. The command takes a device node specifier and a list of r, w or m, separated by a white space.

Example:

root # systemctl set-property system.slice DeviceAllow="/dev/sdb1 r"
DevicePolicy=[auto|closed|strict]

When set to strict, only access to devices that are listed in DeviceAllow is allowed. closed additionally allows access to standard pseudo devices including /dev/null, /dev/zero, /dev/full, /dev/random, and /dev/urandom. auto allows access to all devices if no specific rule is defined in DeviceAllow. auto is the default setting.

For more details and a complete list of properties, see man systemd.resource-control.

9.3 For More Information

10 Automatic Non-Uniform Memory Access (NUMA) Balancing

There are physical limitations to hardware that are encountered when many CPUs and lots of memory are required. In this chapter, the important limitation is that there is limited communication bandwidth between the CPUs and the memory. One architecture modification that was introduced to address this is Non-Uniform Memory Access (NUMA).

In this configuration, there are multiple nodes. Each of the nodes contains a subset of all CPUs and memory. The access speed to main memory is determined by the location of the memory relative to the CPU. The performance of a workload depends on the application threads accessing data that is local to the CPU the thread is executing on. Automatic NUMA Balancing is a new feature of SLE 12. Automatic NUMA Balancing migrates data on demand to memory nodes that are local to the CPU accessing that data. Depending on the workload, this can dramatically boost performance when using NUMA hardware.

10.1 Implementation

Automatic NUMA balancing happens in three basic steps:

  1. A task scanner periodically scans a portion of a task's address space and marks the memory to force a page fault when the data is next accessed.

  2. The next access to the data will result in a NUMA Hinting Fault. Based on this fault, the data can be migrated to a memory node associated with the task accessing the memory.

  3. To keep a task, the CPU it is using and the memory it is accessing together, the scheduler groups tasks that share data.

The unmapping of data and page fault handling incurs overhead. However, commonly the overhead will be offset by threads accessing data associated with the CPU.

10.2 Configuration

Static configuration has been the recommended way of tuning workloads on NUMA hardware for some time. To do this, memory policies can be set with numactl, taskset or cpusets. NUMA-aware applications can use special APIs. In cases where the static policies have already been created, automatic NUMA balancing should be disabled as the data access should already be local.

numactl --hardware will show the memory configuration of the machine and whether it supports NUMA or not. This is example output from a 4-node machine.

tux > numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 4 8 12 16 20 24 28 32 36 40 44
node 0 size: 16068 MB
node 0 free: 15909 MB
node 1 cpus: 1 5 9 13 17 21 25 29 33 37 41 45
node 1 size: 16157 MB
node 1 free: 15948 MB
node 2 cpus: 2 6 10 14 18 22 26 30 34 38 42 46
node 2 size: 16157 MB
node 2 free: 15981 MB
node 3 cpus: 3 7 11 15 19 23 27 31 35 39 43 47
node 3 size: 16157 MB
node 3 free: 16028 MB
node distances:
node   0   1   2   3
  0:  10  20  20  20
  1:  20  10  20  20
  2:  20  20  10  20
  3:  20  20  20  10

Automatic NUMA balancing can be enabled or disabled for the current session by writing 1 or 0 to /proc/sys/kernel/numa_balancing which will enable or disable the feature respectively. To permanently enable or disable it, use the kernel command line option numa_balancing=[enable|disable].

If Automatic NUMA Balancing is enabled, the task scanner behavior can be configured. The task scanner balances the overhead of Automatic NUMA Balancing with the amount of time it takes to identify the best placement of data.

numa_balancing_scan_delay_ms

The amount of CPU time a thread must consume before its data is scanned. This prevents creating overhead because of short-lived processes.

numa_balancing_scan_period_min_ms and numa_balancing_scan_period_max_ms

Controls how frequently a task's data is scanned. Depending on the locality of the faults the scan rate will increase or decrease. These settings control the min and max scan rates.

numa_balancing_scan_size_mb

Controls how much address space is scanned when the task scanner is active.

10.3 Monitoring

The most important task is to assign metrics to your workload and measure the performance with Automatic NUMA Balancing enabled and disabled to measure the impact. Profiling tools can be used to monitor local and remote memory accesses if the CPU supports such monitoring. Automatic NUMA Balancing activity can be monitored via the following parameters in /proc/vmstat:

numa_pte_updates

The amount of base pages that were marked for NUMA hinting faults.

numa_huge_pte_updates

The amount of transparent huge pages that were marked for NUMA hinting faults. In combination with numa_pte_updates the total address space that was marked can be calculated.

numa_hint_faults

Records how many NUMA hinting faults were trapped.

numa_hint_faults_local

Shows how many of the hinting faults were to local nodes. In combination with numa_hint_faults, the percentage of local versus remote faults can be calculated. A high percentage of local hinting faults indicates that the workload is closer to being converged.

numa_pages_migrated

Records how many pages were migrated because they were misplaced. As migration is a copying operation, it contributes the largest part of the overhead created by NUMA balancing.

10.4 Impact

The following illustrates a simple test case of a 4-node NUMA machine running the SpecJBB 2005 using a single instance of the JVM with no static tuning around memory policies. Note, however, that the impact for each workload will vary and that this example is based on a pre-release version of SUSE Linux Enterprise Server 12.

            Balancing disabled      Balancing enabled
TPut 1      26629.00 (  0.00%)     26507.00 ( -0.46%)
TPut 2      55841.00 (  0.00%)     53592.00 ( -4.03%)
TPut 3      86078.00 (  0.00%)     86443.00 (  0.42%)
TPut 4     116764.00 (  0.00%)    113272.00 ( -2.99%)
TPut 5     143916.00 (  0.00%)    141581.00 ( -1.62%)
TPut 6     166854.00 (  0.00%)    166706.00 ( -0.09%)
TPut 7     195992.00 (  0.00%)    192481.00 ( -1.79%)
TPut 8     222045.00 (  0.00%)    227143.00 (  2.30%)
TPut 9     248872.00 (  0.00%)    250123.00 (  0.50%)
TPut 10    270934.00 (  0.00%)    279314.00 (  3.09%)
TPut 11    297217.00 (  0.00%)    301878.00 (  1.57%)
TPut 12    311021.00 (  0.00%)    326048.00 (  4.83%)
TPut 13    324145.00 (  0.00%)    346855.00 (  7.01%)
TPut 14    345973.00 (  0.00%)    378741.00 (  9.47%)
TPut 15    354199.00 (  0.00%)    394268.00 ( 11.31%)
TPut 16    378016.00 (  0.00%)    426782.00 ( 12.90%)
TPut 17    392553.00 (  0.00%)    437772.00 ( 11.52%)
TPut 18    396630.00 (  0.00%)    456715.00 ( 15.15%)
TPut 19    399114.00 (  0.00%)    484020.00 ( 21.27%)
TPut 20    413907.00 (  0.00%)    493618.00 ( 19.26%)
TPut 21    413173.00 (  0.00%)    510386.00 ( 23.53%)
TPut 22    420256.00 (  0.00%)    521016.00 ( 23.98%)
TPut 23    425581.00 (  0.00%)    536214.00 ( 26.00%)
TPut 24    429052.00 (  0.00%)    532469.00 ( 24.10%)
TPut 25    426127.00 (  0.00%)    526548.00 ( 23.57%)
TPut 26    422428.00 (  0.00%)    531994.00 ( 25.94%)
TPut 27    424378.00 (  0.00%)    488340.00 ( 15.07%)
TPut 28    419338.00 (  0.00%)    543016.00 ( 29.49%)
TPut 29    403347.00 (  0.00%)    529178.00 ( 31.20%)
TPut 30    408681.00 (  0.00%)    510621.00 ( 24.94%)
TPut 31    406496.00 (  0.00%)    499781.00 ( 22.95%)
TPut 32    404931.00 (  0.00%)    502313.00 ( 24.05%)
TPut 33    397353.00 (  0.00%)    522418.00 ( 31.47%)
TPut 34    382271.00 (  0.00%)    491989.00 ( 28.70%)
TPut 35    388965.00 (  0.00%)    493012.00 ( 26.75%)
TPut 36    374702.00 (  0.00%)    502677.00 ( 34.15%)
TPut 37    367578.00 (  0.00%)    500588.00 ( 36.19%)
TPut 38    367121.00 (  0.00%)    496977.00 ( 35.37%)
TPut 39    355956.00 (  0.00%)    489430.00 ( 37.50%)
TPut 40    350855.00 (  0.00%)    487802.00 ( 39.03%)
TPut 41    345001.00 (  0.00%)    468021.00 ( 35.66%)
TPut 42    336177.00 (  0.00%)    462260.00 ( 37.50%)
TPut 43    329169.00 (  0.00%)    467906.00 ( 42.15%)
TPut 44    329475.00 (  0.00%)    470784.00 ( 42.89%)
TPut 45    323845.00 (  0.00%)    450739.00 ( 39.18%)
TPut 46    323878.00 (  0.00%)    435457.00 ( 34.45%)
TPut 47    310524.00 (  0.00%)    403914.00 ( 30.07%)
TPut 48    311843.00 (  0.00%)    459017.00 ( 47.19%)

                        Balancing Disabled        Balancing Enabled
 Expctd Warehouse          48.00 (  0.00%)          48.00 (  0.00%)
 Expctd Peak Bops      310524.00 (  0.00%)      403914.00 ( 30.07%)
 Actual Warehouse          25.00 (  0.00%)          29.00 ( 16.00%)
 Actual Peak Bops      429052.00 (  0.00%)      543016.00 ( 26.56%)
 SpecJBB Bops            6364.00 (  0.00%)        9368.00 ( 47.20%)
 SpecJBB Bops/JVM        6364.00 (  0.00%)        9368.00 ( 47.20%)

Automatic NUMA Balancing simplifies tuning workloads for high performance on NUMA machines. Where possible, it is still recommended to statically tune the workload to partition it within each node. However, in all other cases, automatic NUMA balancing should boost performance.

11 Power Management

Power management aims at reducing operating costs for energy and cooling systems while at the same time keeping the performance of a system at a level that matches the current requirements. Thus, power management is always a matter of balancing the actual performance needs and power saving options for a system. Power management can be implemented and used at different levels of the system. A set of specifications for power management functions of devices and the operating system interface to them has been defined in the Advanced Configuration and Power Interface (ACPI). As power savings in server environments can primarily be achieved at the processor level, this chapter introduces some main concepts and highlights some tools for analyzing and influencing relevant parameters.

11.1 Power Management at CPU Level

At the CPU level, you can control power usage in various ways. For example by using idling power states (C-states), changing CPU frequency (P-states), and throttling the CPU (T-states). The following sections give a short introduction to each approach and its significance for power savings. Detailed specifications can be found at http://www.acpi.info/spec.htm.

11.1.1 C-States (Processor Operating States)

Modern processors have several power saving modes called C-states. They reflect the capability of an idle processor to turn off unused components to save power.

When a processor is in the C0 state, it is executing instructions. A processor running in any other C-state is idle. The higher the C number, the deeper the CPU sleep mode: more components are shut down to save power. Deeper sleep states can save large amounts of energy. Their downside is that they introduce latency. This means, it takes more time for the CPU to go back to C0. Depending on workload (threads waking up, triggering CPU usage and then going back to sleep again for a short period of time) and hardware (for example, interrupt activity of a network device), disabling the deepest sleep states can significantly increase overall performance. For details on how to do so, refer to Section 11.3.2, “Viewing Kernel Idle Statistics with cpupower.

Some states also have submodes with different power saving latency levels. Which C-states and submodes are supported depends on the respective processor. However, C1 is always available.

Table 11.1, “C-States” gives an overview of the most common C-states.

Table 11.1: C-States

Mode

Definition

C0

Operational state. CPU fully turned on.

C1

First idle state. Stops CPU main internal clocks via software. Bus interface unit and APIC are kept running at full speed.

C2

Stops CPU main internal clocks via hardware. State in which the processor maintains all software-visible states, but may take longer to wake up through interrupts.

C3

Stops all CPU internal clocks. The processor does not need to keep its cache coherent, but maintains other states. Some processors have variations of the C3 state that differ in how long it takes to wake the processor through interrupts.

To avoid needless power consumption, it is recommended to test your workloads with deep sleep states enabled versus deep sleep states disabled. For more information, refer to Section 11.3.2, “Viewing Kernel Idle Statistics with cpupower or the cpupower-idle-set(1) man page.

11.1.2 P-States (Processor Performance States)

While a processor operates (in C0 state), it can be in one of several CPU performance states (P-states). Whereas C-states are idle states (all but C0), P-states are operational states that relate to CPU frequency and voltage.

The higher the P-state, the lower the frequency and voltage at which the processor runs. The number of P-states is processor-specific and the implementation differs across the various types. However, P0 is always the highest-performance state (except for Section 11.1.3, “Turbo Features”). Higher P-state numbers represent slower processor speeds and lower power consumption. For example, a processor in P3 state runs more slowly and uses less power than a processor running in the P1 state. To operate at any P-state, the processor must be in the C0 state, which means that it is working and not idling. The CPU P-states are also defined in the ACPI specification, see http://www.acpi.info/spec.htm.

C-states and P-states can vary independently of one another.

11.1.3 Turbo Features

Turbo features allow to dynamically overtick active CPU cores while other cores are in deep sleep states. This increases the performance of active threads while still complying with Thermal Design Power (TDP) limits.

However, the conditions under which a CPU core can use turbo frequencies are architecture-specific. Learn how to evaluate the efficiency of those new features in Section 11.3, “The cpupower Tools”.

11.2 In-Kernel Governors

The in-kernel governors belong to the Linux kernel CPUfreq infrastructure and can be used to dynamically scale processor frequencies at runtime. You can think of the governors as a sort of preconfigured power scheme for the CPU. The CPUfreq governors use P-states to change frequencies and lower power consumption. The dynamic governors can switch between CPU frequencies, based on CPU usage, to allow for power savings while not sacrificing performance.

The following governors are available with the CPUfreq subsystem:

Performance Governor

The CPU frequency is statically set to the highest possible for maximum performance. Consequently, saving power is not the focus of this governor.

See also Section 11.4.1, “Tuning Options for P-States”.

Powersave Governor

The CPU frequency is statically set to the lowest possible. This can have severe impact on the performance, as the system will never rise above this frequency no matter how busy the processors are. An important exception is the intel_pstate which defaults to the powersave mode. This is due to a hardware-specific decision but functionally it operates similarly to the on-demand governor.

However, using this governor often does not lead to the expected power savings as the highest savings can usually be achieved at idle through entering C-states. With the powersave governor, processes run at the lowest frequency and thus take longer to finish. This means it takes longer until the system can go into an idle C-state.

Tuning options: The range of minimum frequencies available to the governor can be adjusted (for example, with the cpupower command line tool).

On-demand Governor

The kernel implementation of a dynamic CPU frequency policy: The governor monitors the processor usage. When it exceeds a certain threshold, the governor will set the frequency to the highest available. If the usage is less than the threshold, the next lowest frequency is used. If the system continues to be underemployed, the frequency is again reduced until the lowest available frequency is set.

Important
Important: Drivers and In-kernel Governors

Not all drivers use the in-kernel governors to dynamically scale power frequency at runtime. For example, the intel_pstate driver adjusts power frequency itself. Use the cpupower frequency-info command to find out which driver your system uses.

11.3 The cpupower Tools

The cpupower tools are designed to give an overview of all CPU power-related parameters that are supported on a given machine, including turbo (or boost) states. Use the tool set to view and modify settings of the kernel-related CPUfreq and cpuidle systems and other settings not related to frequency scaling or idle states. The integrated monitoring framework can access both kernel-related parameters and hardware statistics. Therefore, it is ideally suited for performance benchmarks. It also helps you to identify the dependencies between turbo and idle states.

After installing the cpupower package, view the available cpupower subcommands with cpupower --help. Access the general man page with man cpupower, and the man pages of the subcommands with man cpupower-SUBCOMMAND.

11.3.1 Viewing Current Settings with cpupower

The cpupower frequency-info command shows the statistics of the cpufreq driver used in the kernel. Additionally, it shows if turbo (boost) states are supported and enabled in the BIOS. Run without any options, it shows an output similar to the following:

Example 11.1: Example Output of cpupower frequency-info
root # cpupower frequency-info
analyzing CPU 0:
  driver: intel_pstate
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 0.97 ms.
  hardware limits: 1.20 GHz - 3.80 GHz
  available cpufreq governors: performance, powersave
  current policy: frequency should be within 1.20 GHz and 3.80 GHz.
                  The governor "powersave" may decide which speed to use
                  within this range.
  current CPU frequency is 3.40 GHz (asserted by call to hardware).
  boost state support:
    Supported: yes
    Active: yes
    3500 MHz max turbo 4 active cores
    3600 MHz max turbo 3 active cores
    3600 MHz max turbo 2 active cores
    3800 MHz max turbo 1 active cores

To get the current values for all CPUs, use cpupower -c all frequency-info.

11.3.2 Viewing Kernel Idle Statistics with cpupower

The idle-info subcommand shows the statistics of the cpuidle driver used in the kernel. It works on all architectures that use the cpuidle kernel framework.

Example 11.2: Example Output of cpupower idle-info
root # cpupower idle-info
CPUidle driver: intel_idle
CPUidle governor: menu

Analyzing CPU 0:
Number of idle states: 6
Available idle states: POLL C1-SNB C1E-SNB C3-SNB C6-SNB C7-SNB
POLL:
Flags/Description: CPUIDLE CORE POLL IDLE
Latency: 0
Usage: 163128
Duration: 17585669
C1-SNB:
Flags/Description: MWAIT 0x00
Latency: 2
Usage: 16170005
Duration: 697658910
C1E-SNB:
Flags/Description: MWAIT 0x01
Latency: 10
Usage: 4421617
Duration: 757797385
C3-SNB:
Flags/Description: MWAIT 0x10
Latency: 80
Usage: 2135929
Duration: 735042875
C6-SNB:
Flags/Description: MWAIT 0x20
Latency: 104
Usage: 53268
Duration: 229366052
C7-SNB:
Flags/Description: MWAIT 0x30
Latency: 109
Usage: 62593595
Duration: 324631233978

After finding out which processor idle states are supported with cpupower idle-info, individual states can be disabled using the cpupower idle-set command. Typically one wants to disable the deepest sleep state, for example:

cpupower idle-set -d 5

Or, for disabling all CPUs with latencies equal to or higher than 80:

cpupower idle-set -D 80

11.3.3 Monitoring Kernel and Hardware Statistics with cpupower

Use the monitor subcommand to report processor topology, and monitor frequency and idle power state statistics over a certain period of time. The default interval is 1 second, but it can be changed with the -i. Independent processor sleep states and frequency counters are implemented in the tool—some retrieved from kernel statistics, others reading out hardware registers. The available monitors depend on the underlying hardware and the system. List them with cpupower monitor -l. For a description of the individual monitors, refer to the cpupower-monitor man page.

The monitor subcommand allows you to execute performance benchmarks. To compare kernel statistics with hardware statistics for specific workloads, concatenate the respective command, for example:

cpupower monitor db_test.sh
Example 11.3: Example cpupower monitor Output
root # cpupower monitor
|Mperf                   || Idle_Stats
 1                         2 
CPU | C0   | Cx   | Freq || POLL | C1   | C2   | C3
   0|  3.71| 96.29|  2833||  0.00|  0.00|  0.02| 96.32
   1| 100.0| -0.00|  2833||  0.00|  0.00|  0.00|  0.00
   2|  9.06| 90.94|  1983||  0.00|  7.69|  6.98| 76.45
   3|  7.43| 92.57|  2039||  0.00|  2.60| 12.62| 77.52

1

Mperf shows the average frequency of a CPU, including boost frequencies, over time. Additionally, it shows the percentage of time the CPU has been active (C0) or in any sleep state (Cx). As the turbo states are managed by the BIOS, it is impossible to get the frequency values at a given instant. On modern processors with turbo features the Mperf monitor is the only way to find out about the frequency a certain CPU has been running in.

2

Idle_Stats shows the statistics of the cpuidle kernel subsystem. The kernel updates these values every time an idle state is entered or left. Therefore there can be some inaccuracy when cores are in an idle state for some time when the measure starts or ends.

Apart from the (general) monitors in the example above, other architecture-specific monitors are available. For detailed information, refer to the cpupower-monitor man page.

By comparing the values of the individual monitors, you can find correlations and dependencies and evaluate how well the power saving mechanism works for a certain workload. In Example 11.3 you can see that CPU 0 is idle (the value of Cx is near 100%), but runs at a very high frequency. This is because the CPUs 0 and 1 have the same frequency values which means that there is a dependency between them.

11.3.4 Modifying Current Settings with cpupower

You can use cpupower frequency-set command as root to modify current settings. It allows you to set values for the minimum or maximum CPU frequency the governor may select or to create a new governor. With the -c option, you can also specify for which of the processors the settings should be modified. That makes it easy to use a consistent policy across all processors without adjusting the settings for each processor individually. For more details and the available options, refer to the cpupower-freqency-set man page or run cpupower frequency-set --help.

11.4 Special Tuning Options

The following sections highlight important settings.

11.4.1 Tuning Options for P-States

The CPUfreq subsystem offers several tuning options for P-states: You can switch between the different governors, influence minimum or maximum CPU frequency to be used or change individual governor parameters.

To switch to another governor at runtime, use cpupower frequency-set with the -g option. For example, running the following command (as root) will activate the performance governor:

cpupower frequency-set -g performance

To set values for the minimum or maximum CPU frequency the governor may select, use the -d or -u option, respectively.

11.5 Troubleshooting

BIOS options enabled?

To use C-states or P-states, check your BIOS options:

  • To use C-states, make sure to enable CPU C State or similar options to benefit from power savings at idle.

  • To use P-states and the CPUfreq governors, make sure to enable Processor Performance States options or similar.

  • Even if P-states and C-states are available, it is possible that the platform firmware is managing CPU frequencies which may be sub-optimal. For example, if pcc-cpufreq is loaded then the OS is only giving hints to the firmware, which is free to ignore the hints. This can be addressed by selecting "OS Management" or similar for CPU frequency managed in the BIOS. After reboot, an alternative driver will be used but the performance impact should be carefully measured.

In case of a CPU upgrade, make sure to upgrade your BIOS, too. The BIOS needs to know the new CPU and its frequency stepping to pass this information on to the operating system.

Log file information?

Check the systemd journal (see Book “Administration Guide”, Chapter 16 “journalctl: Query the systemd Journal”) for any output regarding the CPUfreq subsystem. Only severe errors are reported there.

If you suspect problems with the CPUfreq subsystem on your machine, you can also enable additional debug output. To do so, either use cpufreq.debug=7 as boot parameter or execute the following command as root:

echo 7 > /sys/module/cpufreq/parameters/debug

This will cause CPUfreq to log more information to dmesg on state transitions, which is useful for diagnosis. But as this additional output of kernel messages can be rather comprehensive, use it only if you are fairly sure that a problem exists.

11.6 For More Information

Platforms with a Baseboard Management Controller (BMC) may have additional power management configuration options accessible via the service processor. These configurations are vendor specific and therefore not subject of this guide. For more information, refer to the manuals provided by your vendor.

11.7 Monitoring Power Consumption with powerTOP

powerTOP helps to identify the causes of unnecessary high power consumption. This is especially useful for laptops, where minimizing power consumption is more important. It supports both Intel and AMD processors. Install it in the usual way:

tux > sudo zypper in powertop

powerTOP combines various sources of information (analysis of programs, device drivers, kernel options, number and sources of interrupts waking up processors from sleep states) and provides several ways of viewing them. You may launch it interactive mode, which runs in an ncurses session (see Figure 11.1, “powerTOP in Interactive Mode”):

tux > sudo powertop
powerTOP in Interactive Mode
Figure 11.1: powerTOP in Interactive Mode

powerTOP supports exporting reports to HTML and CSV. The following example generates a single report of a 240-second run:

tux > sudo powertop --iteration=1 --time=240 --html=POWERREPORT.HTML

It can be useful to run separate reports over time. The following example runs powerTOP 10 times for 20 seconds each time, and creates a separate HTML report for each run:

tux > sudo powertop --iteration=10 --time=20 --html=POWERREPORT.HTML

This creates 10 time-stamped reports:

powerreport-20200108-104512.html
powerreport-20200108-104451.html
powerreport-20200108-104431.html
[...]

An HTML report looks like Figure 11.2, “HTML powerTOP Report”:

HTML powerTOP Report
Figure 11.2: HTML powerTOP Report

The Tuning tab of the HTML reports, and the Tunables tab in the interactive mode, both provide commands for testing the various power settings. The HTML report prints the commands, which you can copy to a root command line for testing, for example echo '0' > '/proc/sys/kernel/nmi_watchdog'. The ncurses mode provides a simple toggle between Good and Bad. Good runs a command to enable power saving, and Bad turns off power saving. Enable all powerTOP settings with one command:

tux > sudo powertop --auto-tune

None of these changes survive a reboot. To make any changes permanent, use sysctl, udev, or systemd to run your selected commands at boot. powerTOP includes a systemd service file, /usr/lib/systemd/system/powertop.service. This starts powerTOP with the --auto-tune option:

ExecStart=/usr/sbin/powertop --auto-tune

Test this carefully before launching the systemd service, to see if it gives the results that you want. You probably do not want USB keyboards and mice to go into powersave modes because it is a nuisance to continually wake them up, and there may other devices you want left alone. For easier testing and configuration editing, extract the commands from an HTML report with awk:

tux > awk -F '</?td ?>' '/tune/ { print $4 }' POWERREPORT.HTML

In calibrate mode, powerTOP sets up several runs that use different idle settings for backlight, CPU, wifi, USB devices, and disks, and helps to identify optimal brightness settings on battery power:

tux > sudo powertop --calibrate

You may call a file that creates a workload for more accurate calibration:

tux > sudo powertop --calibrate --workload=FILENAME --html=POWERREPORT.HTML

For more information, see:

Part V Kernel Tuning

  • 12 Tuning I/O Performance
  • I/O scheduling controls how input/output operations will be submitted to storage. SUSE Linux Enterprise Server offers various I/O algorithms—called elevators—suiting different workloads. Elevators can help to reduce seek operations and can prioritize I/O requests.

  • 13 Tuning the Task Scheduler
  • Modern operating systems, such as SUSE® Linux Enterprise Server, normally run many tasks at the same time. For example, you can be searching in a text file while receiving an e-mail and copying a big file to an external hard disk. These simple tasks require many additional processes to be run by the…

  • 14 Tuning the Memory Management Subsystem
  • To understand and tune the memory management behavior of the kernel, it is important to first have an overview of how it works and cooperates with other subsystems.

  • 15 Tuning the Network
  • The network subsystem is complex and its tuning highly depends on the system use scenario and on external factors such as software clients or hardware components (switches, routers, or gateways) in your network. The Linux kernel aims more at reliability and low latency than low overhead and high thr…

12 Tuning I/O Performance

I/O scheduling controls how input/output operations will be submitted to storage. SUSE Linux Enterprise Server offers various I/O algorithms—called elevators—suiting different workloads. Elevators can help to reduce seek operations and can prioritize I/O requests.

Choosing the best suited I/O elevator not only depends on the workload, but on the hardware, too. Single ATA disk systems, SSDs, RAID arrays, or network storage systems, for example, each require different tuning strategies.

12.1 Switching I/O Scheduling

SUSE Linux Enterprise Server picks a default I/O scheduler at boot-time, which can be changed on the fly per block device. This makes it possible to set different algorithms, for example, for the device hosting the system partition and the device hosting a database.

The default I/O scheduler is chosen for each device based on whether the device reports to be rotational disk or not. For non-rotational disks DEADLINE I/O scheduler is picked. Other devices default to CFQ (Completely Fair Queuing). To change this default, use the following boot parameter:

elevator=SCHEDULER

Replace SCHEDULER with one of the values cfq, noop, or deadline. See Section 12.2, “Available I/O Elevators” for details.

To change the elevator for a specific device in the running system, run the following command:

echo SCHEDULER > /sys/block/DEVICE/queue/scheduler

Here, SCHEDULER is one of cfq, noop, or deadline. DEVICE is the block device (sda for example). Note that this change will not persist during reboot. For permanent I/O scheduler change for a particular device either place the command switching the I/O scheduler into init scripts or add appropriate udev rule into /lib/udev/rules.d/. See /lib/udev/rules.d/60-ssd-scheduler.rules for an example of such tuning.

Note
Note: Default Scheduler on IBM IBM Z

On IBM IBM Z, the default I/O scheduler for a storage device is set by the device driver.

Note
Note: Scheduler in case of block multi-queue (blk-mq) I/O path

The elevator boot parameter does not apply to devices using blk-mq I/O path (refer to Section 12.5, “Enable blk-mq I/O Path for SCSI by Default”).

Default elevator is mq-deadline for conventional devices (for example, regular hard disks, SSDs) and none for faster storage devices (devices with multiple hardware queues).

It is possible to change the elevator for a specific device in a running system. SCHEDULER should be set to either mq-deadline, kyber, bfq, or none

12.2 Available I/O Elevators

Below is a list of elevators available on SUSE Linux Enterprise Server for devices that use the legacy block I/O path. If an elevator has tunable parameters, they can be set with the command:

echo VALUE > /sys/block/DEVICE/queue/iosched/TUNABLE

where VALUE is the desired value for the TUNABLE and DEVICE the block device.

To find out what elevators are available for a device (sda for example), run the following command (the currently selected scheduler is listed in brackets):

jupiter:~ # cat /sys/block/sda/queue/scheduler
noop deadline [cfq]

This file can also contain the string none meaning that I/O scheduling does not happen for this device. This is usually because the device uses a multi-queue queuing mechanism (refer to Section 12.5, “Enable blk-mq I/O Path for SCSI by Default”).

12.2.1 CFQ (Completely Fair Queuing)

CFQ is a fairness-oriented scheduler and is used by default on SUSE Linux Enterprise Server. The algorithm assigns each thread a time slice in which it is allowed to submit I/O to disk. This way each thread gets a fair share of I/O throughput. It also allows assigning tasks I/O priorities which are taken into account during scheduling decisions (see Section 8.3.3, “Prioritizing Disk Access with ionice). The CFQ scheduler has the following tunable parameters:

Table 12.1: CFQ tunable parameters

File

Description

slice_idle

When a task has no more I/O to submit in its time slice, the I/O scheduler waits before scheduling the next thread. slice_idle specifies the I/O scheduler's waiting time in milliseconds. Waiting for more I/O from a thread can improve locality of I/O. Additionally, it avoids starving processes doing dependent I/O. A process does dependent I/O if it needs a result of one I/O to submit another I/O. For example, if you first need to read an index block to find out a data block to read, these two reads form a dependent I/O.

For media where locality is less important (SSDs, SANs with lots of disks), setting slice_idle to 0 can improve the throughput considerably.

Default is 8.

slice_idle_us

Same as slice_idle but in microseconds.

Default is 8000.

quantum

This option limits the maximum number of requests that are being processed by the device. For a storage with several disks, this setting can unnecessarily limit parallel processing of requests. Therefore, increasing the value can improve performance. However, it can also cause latency of certain I/O operations to increase, because more requests are buffered inside the storage. When changing this value, you can also consider tuning slice_async_rq.

Default is 8.

low_latency

When enabled (which is the default on SUSE Linux Enterprise Server), the scheduler may dynamically adjust the length of the time slice by aiming to meet a tuning parameter called the target_latency. Time slices are recomputed to meet this target_latency and ensure that processes get fair access within a bounded length of time.

Default is 1.

target_latency

Contains an estimated latency time in milliseconds for CFQ. CFQ uses it to calculate the time slice used for every task.

Default is 300.

target_latency_us

Same as target_latency but in microseconds.

Default is 300000.

group_idle

To avoid starving of blkio cgroups doing dependent I/O, CFQ pauses after completion of I/O for one blkio cgroup before scheduling I/O for a different blkio cgroup. group_idle specifies the time in milliseconds the I/O scheduler waits. When slice_idle is set, this parameter does not have a significant effect. However, for fast media, the overhead of slice_idle is generally undesirable. Disabling slice_idle and setting group_idle is a method to avoid starvation of blkio cgroups doing dependent I/O with lower overhead.

Default is 8.

group_idle_us

Same as group_idle but in microseconds.

Default is 8000.

slice_sync

This parameter is used to calculate the time slice for synchronous queue. It is specified in milliseconds. Increasing this value increases the time slice of synchronous queue.

Default is 100.

slice_sync_us

Same as slice_sync but in microseconds.

Default is 100000.

slice_async

This parameter is used to calculate the time slice for asynchronous queue. It is specified in milliseconds. Increasing this value increases the time slice of asynchronous queue.

Default is 40.

slice_async_us

Same as slice_async but in microseconds.

Default is 40000.

slice_async_rq

This limits the maximum number of asynchronous requests—usually write requests—that are submitted in one time slice.

Default is 2.

back_seek_max

Maximum "distance" (in Kbytes) for backward seeking.

Default is 16384.

back_seek_penalty

Used to compute the cost of backward seeking.

Default is 2.

fifo_expire_async

Value (in milliseconds) is used to set the timeout of asynchronous requests.

Default is 250.

fifo_expire_sync

Value (in milliseconds) that specifies the timeout of synchronous requests.

Default is 125.

Example 12.1: Increasing individual thread throughput using CFQ

In SUSE Linux Enterprise Server 12 SP5, the low_latency tuning parameter is enabled by default to ensure that processes get fair access within a bounded length of time. (Note that this parameter was not enabled in versions prior to SUSE Linux Enterprise 12.)

This is usually preferred in a server scenario where processes are executing I/O as part of transactions, as it makes the time needed for each transaction predictable. However, there are scenarios where that is not the desired behavior:

  • If the performance metric of interest is the peak performance of a single process when there is I/O contention.

  • If a workload must complete as quickly as possible and there are multiple sources of I/O. In this case, unfair treatment from the I/O scheduler may allow the transactions to complete faster: Processes take their full slice and exit quickly, resulting in reduced overall contention.

To address this, there are two options—increase target_latency or disable low_latency. As with all tuning parameters it is important to verify your workload behaves as expected before and after the tuning modification. Take careful note of whether your workload depends on individual process peak performance or scales better with fairness. It should also be noted that the performance will depend on the underlying storage and the correct tuning option for one installation may not be universally true.

Find below an example that does not control when I/O starts but is simple enough to demonstrate the point. 32 processes are writing a small amount of data to disk in parallel. Using the SUSE Linux Enterprise Server default (enabling low_latency), the result looks as follows:

root # echo 1 > /sys/block/sda/queue/iosched/low_latency
root # time ./dd-test.sh
10485760 bytes (10 MB) copied, 2.62464 s, 4.0 MB/s
10485760 bytes (10 MB) copied, 3.29624 s, 3.2 MB/s
10485760 bytes (10 MB) copied, 3.56341 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.56908 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.53043 s, 3.0 MB/s
10485760 bytes (10 MB) copied, 3.57511 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.53672 s, 3.0 MB/s
10485760 bytes (10 MB) copied, 3.5433 s, 3.0 MB/s
10485760 bytes (10 MB) copied, 3.65474 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.63694 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.90122 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 3.88507 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 3.86135 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 3.84553 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 3.88871 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 3.94943 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 4.12731 s, 2.5 MB/s
10485760 bytes (10 MB) copied, 4.15106 s, 2.5 MB/s
10485760 bytes (10 MB) copied, 4.21601 s, 2.5 MB/s
10485760 bytes (10 MB) copied, 4.35004 s, 2.4 MB/s
10485760 bytes (10 MB) copied, 4.33387 s, 2.4 MB/s
10485760 bytes (10 MB) copied, 4.55434 s, 2.3 MB/s
10485760 bytes (10 MB) copied, 4.52283 s, 2.3 MB/s
10485760 bytes (10 MB) copied, 4.52682 s, 2.3 MB/s
10485760 bytes (10 MB) copied, 4.56176 s, 2.3 MB/s
10485760 bytes (10 MB) copied, 4.62727 s, 2.3 MB/s
10485760 bytes (10 MB) copied, 4.78958 s, 2.2 MB/s
10485760 bytes (10 MB) copied, 4.79772 s, 2.2 MB/s
10485760 bytes (10 MB) copied, 4.78004 s, 2.2 MB/s
10485760 bytes (10 MB) copied, 4.77994 s, 2.2 MB/s
10485760 bytes (10 MB) copied, 4.86114 s, 2.2 MB/s
10485760 bytes (10 MB) copied, 4.88062 s, 2.1 MB/s

real    0m4.978s
user    0m0.112s
sys     0m1.544s

Note that each process completes in similar times. This is the CFQ scheduler meeting its target_latency: Each process has fair access to storage.

Note that the earlier processes complete somewhat faster. This happens because the start time of the processes is not identical. In a more complicated example, it is possible to control for this.

This is what happens when low_latency is disabled:

root # echo 0 > /sys/block/sda/queue/iosched/low_latency
root # time ./dd-test.sh
10485760 bytes (10 MB) copied, 0.813519 s, 12.9 MB/s
10485760 bytes (10 MB) copied, 0.788106 s, 13.3 MB/s
10485760 bytes (10 MB) copied, 0.800404 s, 13.1 MB/s
10485760 bytes (10 MB) copied, 0.816398 s, 12.8 MB/s
10485760 bytes (10 MB) copied, 0.959087 s, 10.9 MB/s
10485760 bytes (10 MB) copied, 1.09563 s, 9.6 MB/s
10485760 bytes (10 MB) copied, 1.18716 s, 8.8 MB/s
10485760 bytes (10 MB) copied, 1.27661 s, 8.2 MB/s
10485760 bytes (10 MB) copied, 1.46312 s, 7.2 MB/s
10485760 bytes (10 MB) copied, 1.55489 s, 6.7 MB/s
10485760 bytes (10 MB) copied, 1.64277 s, 6.4 MB/s
10485760 bytes (10 MB) copied, 1.78196 s, 5.9 MB/s
10485760 bytes (10 MB) copied, 1.87496 s, 5.6 MB/s
10485760 bytes (10 MB) copied, 1.9461 s, 5.4 MB/s
10485760 bytes (10 MB) copied, 2.08351 s, 5.0 MB/s
10485760 bytes (10 MB) copied, 2.28003 s, 4.6 MB/s
10485760 bytes (10 MB) copied, 2.42979 s, 4.3 MB/s
10485760 bytes (10 MB) copied, 2.54564 s, 4.1 MB/s
10485760 bytes (10 MB) copied, 2.6411 s, 4.0 MB/s
10485760 bytes (10 MB) copied, 2.75171 s, 3.8 MB/s
10485760 bytes (10 MB) copied, 2.86162 s, 3.7 MB/s
10485760 bytes (10 MB) copied, 2.98453 s, 3.5 MB/s
10485760 bytes (10 MB) copied, 3.13723 s, 3.3 MB/s
10485760 bytes (10 MB) copied, 3.36399 s, 3.1 MB/s
10485760 bytes (10 MB) copied, 3.60018 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.58151 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.67385 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.69471 s, 2.8 MB/s
10485760 bytes (10 MB) copied, 3.66658 s, 2.9 MB/s
10485760 bytes (10 MB) copied, 3.81495 s, 2.7 MB/s
10485760 bytes (10 MB) copied, 4.10172 s, 2.6 MB/s
10485760 bytes (10 MB) copied, 4.0966 s, 2.6 MB/s

real    0m3.505s
user    0m0.160s
sys     0m1.516s

Note that the time processes take to complete is spread much wider as processes are not getting fair access. Some processes complete faster and exit, allowing the total workload to complete faster, and some processes measure higher apparent I/O performance. It is also important to note that this example may not behave similarly on all systems as the results depend on the resources of the machine and the underlying storage.

It is important to emphasize that neither tuning option is inherently better than the other. Both are best in different circumstances and it is important to understand the requirements of your workload and tune accordingly.

12.2.2 NOOP

A trivial scheduler that only passes down the I/O that comes to it. Useful for checking whether complex I/O scheduling decisions of other schedulers are causing I/O performance regressions.

This scheduler is recommended for setups with devices that do I/O scheduling themselves, such as intelligent storage or in multipathing environments. If you choose a more complicated scheduler on the host, the scheduler of the host and the scheduler of the storage device compete with each other. This can decrease performance. The storage device can usually determine best how to schedule I/O.

For similar reasons, this scheduler is also recommended for use within virtual machines.

The NOOP scheduler can be useful for devices that do not depend on mechanical movement, like SSDs. Usually, the DEADLINE I/O scheduler is a better choice for these devices. However, NOOP creates less overhead and thus can on certain workloads increase performance.

12.2.3 DEADLINE

DEADLINE is a latency-oriented I/O scheduler. Each I/O request is assigned a deadline. Usually, requests are stored in queues (read and write) sorted by sector numbers. The DEADLINE algorithm maintains two additional queues (read and write) in which requests are sorted by deadline. As long as no request has timed out, the sector queue is used. When timeouts occur, requests from the deadline queue are served until there are no more expired requests. Generally, the algorithm prefers reads over writes.

This scheduler can provide a superior throughput over the CFQ I/O scheduler in cases where several threads read and write and fairness is not an issue. For example, for several parallel readers from a SAN and for databases (especially when using TCQ disks). The DEADLINE scheduler has the following tunable parameters:

Table 12.2: DEADLINE tunable parameters

File

Description

writes_starved

Controls how many times reads are preferred over writes. A value of 3 means that three read operations can be done before writes and reads are dispatched on the same selection criteria.

Default is 3.

read_expire

Sets the deadline (current time plus the read_expire value) for read operations in milliseconds.

Default is 500.

write_expire

Sets the deadline (current time plus the write_expire value) for write operations in milliseconds.

Default is 5000.

front_merges

Enables (1) or disables (0) attempts to front merge requests.

Default is 1.

fifo_batch

Sets the maximum number of requests per batch (deadline expiration is only checked for batches). This parameter allows to balance between latency and throughput. When set to 1 (that is, one request per batch), it results in "first come, first served" behaviour and usually lowest latency. Higher values usually increase throughput.

Default is 16.

12.3 Available I/O Elevators with blk-mq I/O path

Below is a list of elevators available on SUSE Linux Enterprise Server for devices that use the blk-mq I/O path If an elevator has tunable parameters, they can be set with the command:

echo VALUE > /sys/block/DEVICE/queue/iosched/TUNABLE

In command above, VALUE is the desired value for the TUNABLE and DEVICE is the block device.

To find out what elevators are available for a device (sda for example), run the following command (the currently selected scheduler is listed in brackets):

tux > cat /sys/block/sda/queue/scheduler
[mq-deadline] kyber bfq none
Note
Note: Scheduler options when switching from legacy block to blk-mq I/O path

When switching from legacy block to blk-mq I/O path for a device, the none option is roughly comparable to noop, mq-deadline is comparable to deadline, and bfq is comparable to cfq.

12.3.1 MQ-DEADLINE

MQ-DEADLINE is a latency-oriented I/O scheduler. It is a modification of DEADLINE scheduler for blk-mq I/O path (refer to Section 12.2.3, “DEADLINE). MQ-DEADLINE has the same set of tunable parameters. Please refer to Table 12.2, “DEADLINE tunable parameters” for a description.

12.3.2 NONE

When NONE is selected as I/O elevator option for blk-mq, no I/O scheduler is used, and I/O requests are passed down to the device without further I/O scheduling interaction. In this respect, it is comparable to NOOP scheduler for the legacy block I/O path (see Section 12.2.2, “NOOP).

NONE is the default for NVM Express devices. With no overhead compared to other I/O elevator options, it is considered the fastest way of passing down I/O requests on multiple queues to such devices.

There are no tunable parameters for NONE.

12.3.3 BFQ (Budget Fair Queueing)

BFQ is a fairness-oriented scheduler. It is described as "a proportional-share storage-I/O scheduling algorithm based on the slice-by-slice service scheme of CFQ. But BFQ assigns budgets, measured in number of sectors, to processes instead of time slices." (Source: linux-4.12/block/bfq-iosched.c)

BFQ allows to assign I/O priorities to tasks which are taken into account during scheduling decisions (see Section 8.3.3, “Prioritizing Disk Access with ionice).

BFQ scheduler has following tunable parameters:

Table 12.3: BFQ tunable parameters

File

Description

slice_idle

Value in milliseconds specifies how long to idle, waiting for next request on an empty queue.

Default is 8.

slice_idle_us

Same as slice_idle but in microseconds.

Default is 8000.

low_latency

Enables (1) or disables (0) BFQ's low latency mode. This mode prioritizes certain applications (for example, if interactive) such that they observe lower latency.

Default is 1.

back_seek_max

Maximum value (in Kbytes) for backward seeking.

Default is 16384.

back_seek_penalty

Used to compute the cost of backward seeking.

Default is 2.

fifo_expire_async

Value (in milliseconds) is used to set the timeout of asynchronous requests.

Default is 250.

fifo_expire_sync

Value in milliseconds specifies the timeout of synchronous requests.

Default is 125.

timeout_sync

Maximum time in milliseconds that a task (queue) is serviced after it has been selected.

Default is 124.

max_budget

Limit for number of sectors that are served at maximum within timeout_sync. If set to 0 BFQ internally calculates a value based on timeout_sync and an estimated peak rate.

Default is 0 (set to auto-tuning).

strict_guarantees

Enables (1) or disables (0) BFQ specific queue handling required to give stricter bandwidth sharing guarantees under certain conditions.

Default is 0.

12.3.4 KYBER

KYBER is a latency-oriented I/O scheduler. It makes it possible to set target latencies for reads and synchronous writes and throttles I/O requests in order to try to meet these target latencies.

Table 12.4: KYBER tunable parameters

File

Description

read_lat_nsec

Sets the target latency for read operations in nanoseconds.

Default is 2000000.

write_lat_nsec

Sets the target latency for write operations in nanoseconds.

Default is 10000000.

12.4 I/O Barrier Tuning

Most file systems (such as XFS, Ext3, Ext4, or ReiserFS) send write barriers to disk after fsync or during transaction commits. Write barriers enforce proper ordering of writes, making volatile disk write caches safe to use (at some performance penalty). If your disks are battery-backed in one way or another, disabling barriers can safely improve performance.

Sending write barriers can be disabled using the nobarrier mount option.

Warning
Warning: Disabling Barriers Can Lead to Data Loss

Disabling barriers when disks cannot guarantee caches are properly written in case of power failure can lead to severe file system corruption and data loss.

12.5 Enable blk-mq I/O Path for SCSI by Default

Block multiqueue (blk-mq) is a multi-queue block I/O queueing mechanism. Blk-mq uses per-cpu software queues to queue I/O requests. The software queues are mapped to one or more hardware submission queues. Blk-mq significantly reduces lock contention. In particular blk-mq improves performance for devices that support a high number of input/output operations per second (IOPS). Blk-mq is already the default for some devices, for example, NVM Express devices.

Blk-mq has a different set of I/O scheduler options. There is MQ-DEADLINE (comparable to DEADLINE) and NONE (comparable to NOOP). There is no longer CFQ I/O scheduler with blk-mq. But there are two new I/O schedulers: BFQ and KYBER. These changes in I/O scheduling can cause performance differences with blk-mq compared to legacy block I/O path. Therefore, blk-mq is not enabled by default for SCSI devices.

If you have fast SCSI devices (for example, SSDs) instead of SCSI hard disks attached to your system, consider switching to blk-mq for SCSI. This is done using the kernel command line option scsi_mod.use_blk_mq=1. If you have also attached SCSI hard disks (spinning devices) to your system, make sure to switch to BFQ I/O scheduler for the spinning devices to avoid their significant performance degradation.

13 Tuning the Task Scheduler

Modern operating systems, such as SUSE® Linux Enterprise Server, normally run many tasks at the same time. For example, you can be searching in a text file while receiving an e-mail and copying a big file to an external hard disk. These simple tasks require many additional processes to be run by the system. To provide each task with its required system resources, the Linux kernel needs a tool to distribute available system resources to individual tasks. And this is exactly what the task scheduler does.

The following sections explain the most important terms related to a process scheduling. They also introduce information about the task scheduler policy, scheduling algorithm, description of the task scheduler used by SUSE Linux Enterprise Server, and references to other sources of relevant information.

13.1 Introduction

The Linux kernel controls the way that tasks (or processes) are managed on the system. The task scheduler, sometimes called process scheduler, is the part of the kernel that decides which task to run next. It is responsible for best using system resources to guarantee that multiple tasks are being executed simultaneously. This makes it a core component of any multitasking operating system.

13.1.1 Preemption

The theory behind task scheduling is very simple. If there are runnable processes in a system, at least one process must always be running. If there are more runnable processes than processors in a system, not all the processes can be running all the time.

Therefore, some processes need to be stopped temporarily, or suspended, so that others can be running again. The scheduler decides what process in the queue will run next.

As already mentioned, Linux, like all other Unix variants, is a multitasking operating system. That means that several tasks can be running at the same time. Linux provides a so called preemptive multitasking, where the scheduler decides when a process is suspended. This forced suspension is called preemption. All Unix flavors have been providing preemptive multitasking since the beginning.

13.1.2 Timeslice

The time period for which a process will be running before it is preempted is defined in advance. It is called a timeslice of a process and represents the amount of processor time that is provided to each process. By assigning timeslices, the scheduler makes global decisions for the running system, and prevents individual processes from dominating over the processor resources.

13.1.3 Process Priority

The scheduler evaluates processes based on their priority. To calculate the current priority of a process, the task scheduler uses complex algorithms. As a result, each process is given a value according to which it is allowed to run on a processor.

13.2 Process Classification

Processes are usually classified according to their purpose and behavior. Although the borderline is not always clearly distinct, generally two criteria are used to sort them. These criteria are independent and do not exclude each other.

One approach is to classify a process either I/O-bound or processor-bound.

I/O-bound

I/O stands for Input/Output devices, such as keyboards, mice, or optical and hard disks. I/O-bound processes spend the majority of time submitting and waiting for requests. They are run very frequently, but for short time intervals, not to block other processes waiting for I/O requests.

processor-bound

On the other hand, processor-bound tasks use their time to execute a code, and usually run until they are preempted by the scheduler. They do not block processes waiting for I/O requests, and, therefore, can be run less frequently but for longer time intervals.

Another approach is to divide processes by type into interactive, batch, and real-time processes.

  • Interactive processes spend a lot of time waiting for I/O requests, such as keyboard or mouse operations. The scheduler must wake up such processes quickly on user request, or the user will find the environment unresponsive. The typical delay is approximately 100 ms. Office applications, text editors or image manipulation programs represent typical interactive processes.

  • Batch processes often run in the background and do not need to be responsive. They usually receive lower priority from the scheduler. Multimedia converters, database search engines, or log files analyzers are typical examples of batch processes.

  • Real-time processes must never be blocked by low-priority processes, and the scheduler guarantees a short response time to them. Applications for editing multimedia content are a good example here.

13.3 Completely Fair Scheduler

Since the Linux kernel version 2.6.23, a new approach has been taken to the scheduling of runnable processes. Completely Fair Scheduler (CFS) became the default Linux kernel scheduler. Since then, important changes and improvements have been made. The information in this chapter applies to SUSE Linux Enterprise Server with kernel version 2.6.32 and higher (including 3.x kernels). The scheduler environment was divided into several parts, and three main new features were introduced:

Modular Scheduler Core

The core of the scheduler was enhanced with scheduling classes. These classes are modular and represent scheduling policies.

Completely Fair Scheduler

Introduced in kernel 2.6.23 and extended in 2.6.24, CFS tries to assure that each process obtains its fair share of the processor time.

Group Scheduling

For example, if you split processes into groups according to which user is running them, CFS tries to provide each of these groups with the same amount of processor time.

As a result, CFS brings optimized scheduling for both servers and desktops.

13.3.1 How CFS Works

CFS tries to guarantee a fair approach to each runnable task. To find the most balanced way of task scheduling, it uses the concept of red-black tree. A red-black tree is a type of self-balancing data search tree which provides inserting and removing entries in a reasonable way so that it remains well balanced. For more information, see the wiki pages of Red-black tree.

When CFS schedules a task it accumulates virtual runtime or vruntime. The next task picked to run is always the task with the minimum accumulated vruntime so far. By balancing the red-black tree when tasks are inserted into the run queue (a planned time line of processes to be executed next), the task with the minimum vruntime is always the first entry in the red-black tree.

The amount of vruntime a task accrues is related to its priority. High priority tasks gain vruntime at a slower rate than low priority tasks, which results in high priority tasks being picked to run on the processor more often.

13.3.2 Grouping Processes

Since the Linux kernel version 2.6.24, CFS can be tuned to be fair to groups rather than to tasks only. Runnable tasks are then grouped to form entities, and CFS tries to be fair to these entities instead of individual runnable tasks. The scheduler also tries to be fair to individual tasks within these entities.

The kernel scheduler lets you group runnable tasks using control groups. For more information, see Chapter 9, Kernel Control Groups.

13.3.3 Kernel Configuration Options

Basic aspects of the task scheduler behavior can be set through the kernel configuration options. Setting these options is part of the kernel compilation process. Because kernel compilation process is a complex task and out of this document's scope, refer to relevant source of information.

Warning
Warning: Kernel Compilation

If you run SUSE Linux Enterprise Server on a kernel that was not shipped with it, for example on a self-compiled kernel, you lose the entire support entitlement.

13.3.4 Terminology

Documents regarding task scheduling policy often use several technical terms which you need to know to understand the information correctly. Here are some:

Latency

Delay between the time a process is scheduled to run and the actual process execution.

Granularity

The relation between granularity and latency can be expressed by the following equation:

gran = ( lat / rtasks ) - ( lat / rtasks / rtasks )

where gran stands for granularity, lat stand for latency, and rtasks is the number of running tasks.

13.3.4.1 Scheduling Policies

The Linux kernel supports the following scheduling policies:

SCHED_FIFO

Scheduling policy designed for special time-critical applications. It uses the First In-First Out scheduling algorithm.

SCHED_BATCH

Scheduling policy designed for CPU-intensive tasks.

SCHED_IDLE

Scheduling policy intended for very low prioritized tasks.

SCHED_OTHER

Default Linux time-sharing scheduling policy used by the majority of processes.

SCHED_RR

Similar to SCHED_FIFO, but uses the Round Robin scheduling algorithm.

13.3.5 Changing Real-time Attributes of Processes with chrt

The chrt command sets or retrieves the real-time scheduling attributes of a running process, or runs a command with the specified attributes. You can get or retrieve both the scheduling policy and priority of a process.

In the following examples, a process whose PID is 16244 is used.

To retrieve the real-time attributes of an existing task:

root # chrt -p 16244
pid 16244's current scheduling policy: SCHED_OTHER
pid 16244's current scheduling priority: 0

Before setting a new scheduling policy on the process, you need to find out the minimum and maximum valid priorities for each scheduling algorithm:

root # chrt -m
SCHED_SCHED_OTHER min/max priority : 0/0
SCHED_SCHED_FIFO min/max priority : 1/99
SCHED_SCHED_RR min/max priority : 1/99
SCHED_SCHED_BATCH min/max priority : 0/0
SCHED_SCHED_IDLE min/max priority : 0/0

In the above example, SCHED_OTHER, SCHED_BATCH, SCHED_IDLE polices only allow for priority 0, while that of SCHED_FIFO and SCHED_RR can range from 1 to 99.

To set SCHED_BATCH scheduling policy:

root # chrt -b -p 0 16244
pid 16244's current scheduling policy: SCHED_BATCH
pid 16244's current scheduling priority: 0

For more information on chrt, see its man page (man 1 chrt).

13.3.6 Runtime Tuning with sysctl

The sysctl interface for examining and changing kernel parameters at runtime introduces important variables by means of which you can change the default behavior of the task scheduler. The syntax of the sysctl is simple, and all the following commands must be entered on the command line as root.

To read a value from a kernel variable, enter

sysctl VARIABLE

To assign a value, enter

sysctl VARIABLE=VALUE

To get a list of all scheduler related sysctl variables, enter

sysctl -A | grep "sched" | grep -v"domain"
root # sysctl -A | grep "sched" | grep -v "domain"
kernel.sched_cfs_bandwidth_slice_us = 5000
kernel.sched_child_runs_first = 0
kernel.sched_compat_yield = 0
kernel.sched_latency_ns = 24000000
kernel.sched_migration_cost_ns = 500000
kernel.sched_min_granularity_ns = 8000000
kernel.sched_nr_migrate = 32
kernel.sched_rr_timeslice_ms = 25
kernel.sched_rt_period_us = 1000000
kernel.sched_rt_runtime_us = 950000
kernel.sched_schedstats = 0
kernel.sched_shares_window_ns = 10000000
kernel.sched_time_avg_ms = 1000
kernel.sched_tunable_scaling = 1
kernel.sched_wakeup_granularity_ns = 10000000

Note that variables ending with _ns and _us accept values in nanoseconds and microseconds, respectively.

A list of the most important task scheduler sysctl tuning variables (located at /proc/sys/kernel/) with a short description follows:

sched_cfs_bandwidth_slice_us

When CFS bandwidth control is in use, this parameter controls the amount of run-time (bandwidth) transferred to a run queue from the task's control group bandwidth pool. Small values allow the global bandwidth to be shared in a fine-grained manner among tasks, larger values reduce transfer overhead. See https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.txt.

sched_child_runs_first

A freshly forked child runs before the parent continues execution. Setting this parameter to 1 is beneficial for an application in which the child performs an execution after fork. For example make -j<NO_CPUS> performs better when sched_child_runs_first is turned off. The default value is 0.

sched_compat_yield

Enables the aggressive CPU yielding behavior of the old O(1) scheduler by moving the relinquishing task to the end of the runnable queue (right-most position in the red-black tree). Applications that depend on the sched_yield(2) syscall behavior may see performance improvements by giving other processes a chance to run when there are highly contended resources (such as locks). On the other hand, given that this call occurs in context switching, misusing the call can hurt the workload. Only use it when you see a drop in performance. The default value is 0.

sched_migration_cost_ns

Amount of time after the last execution that a task is considered to be cache hot in migration decisions. A hot task is less likely to be migrated to another CPU, so increasing this variable reduces task migrations. The default value is 500000 (ns).

If the CPU idle time is higher than expected when there are runnable processes, try reducing this value. If tasks bounce between CPUs or nodes too often, try increasing it.

sched_latency_ns

Targeted preemption latency for CPU bound tasks. Increasing this variable increases a CPU bound task's timeslice. A task's timeslice is its weighted fair share of the scheduling period:

timeslice = scheduling period * (task's weight/total weight of tasks in the run queue)

The task's weight depends on the task's nice level and the scheduling policy. Minimum task weight for a SCHED_OTHER task is 15, corresponding to nice 19. The maximum task weight is 88761, corresponding to nice -20.

Timeslices become smaller as the load increases. When the number of runnable tasks exceeds sched_latency_ns/sched_min_granularity_ns, the slice becomes number_of_running_tasks * sched_min_granularity_ns. Prior to that, the slice is equal to sched_latency_ns.

This value also specifies the maximum amount of time during which a sleeping task is considered to be running for entitlement calculations. Increasing this variable increases the amount of time a waking task may consume before being preempted, thus increasing scheduler latency for CPU bound tasks. The default value is 6000000 (ns).

sched_min_granularity_ns

Minimal preemption granularity for CPU bound tasks. See sched_latency_ns for details. The default value is 4000000 (ns).

sched_wakeup_granularity_ns

The wake-up preemption granularity. Increasing this variable reduces wake-up preemption, reducing disturbance of compute bound tasks. Lowering it improves wake-up latency and throughput for latency critical tasks, particularly when a short duty cycle load component must compete with CPU bound components. The default value is 2500000 (ns).

Warning
Warning: Setting the Right Wake-up Granularity Value

Settings larger than half of sched_latency_ns will result in no wake-up preemption. Short duty cycle tasks will be unable to compete with CPU hogs effectively.

sched_rr_timeslice_ms

Quantum that SCHED_RR tasks are allowed to run before they are preempted and put to the end of the task list.

sched_rt_period_us

Period over which real-time task bandwidth enforcement is measured. The default value is 1000000 (µs).

sched_rt_runtime_us

Quantum allocated to real-time tasks during sched_rt_period_us. Setting to -1 disables RT bandwidth enforcement. By default, RT tasks may consume 95%CPU/sec, thus leaving 5%CPU/sec or 0.05s to be used by SCHED_OTHER tasks. The default value is 950000 (µs).

sched_nr_migrate

Controls how many tasks can be migrated across processors for load-balancing purposes. Because balancing iterates the runqueue with interrupts disabled (softirq), it can incur in irq-latency penalties for real-time tasks. Therefore increasing this value may give a performance boost to large SCHED_OTHER threads at the expense of increased irq-latencies for real-time tasks. The default value is 32.

sched_time_avg_ms

This parameter sets the period over which the time spent running real-time tasks is averaged. That average assists CFS in making load-balancing decisions and gives an indication of how busy a CPU is with high-priority real-time tasks.

The optimal setting for this parameter is highly workload dependent and depends, among other things, on how frequently real-time tasks are running and for how long.

13.3.7 Debugging Interface and Scheduler Statistics

CFS comes with a new improved debugging interface, and provides runtime statistics information. Relevant files were added to the /proc file system, which can be examined simply with the cat or less command. A list of the related /proc files follows with their short description:

/proc/sched_debug

Contains the current values of all tunable variables (see Section 13.3.6, “Runtime Tuning with sysctl) that affect the task scheduler behavior, CFS statistics, and information about the run queues (CFS, RT and deadline) on all available processors. A summary of the task running on each processor is also shown, with the task name and PID, along with scheduler specific statistics. The first being tree-key column, it indicates the task's virtual runtime, and its name comes from the kernel sorting all runnable tasks by this key in a red-black tree. The switches column indicates the total number of switches (involuntary or not), and naturally the prio refers to the process priority. The wait-time value indicates the amount of time the task waited to be scheduled. Finally both sum-exec and sum-sleep account for the total amount of time (in nanoseconds) the task was running on the processor or asleep, respectively.

root # cat /proc/sched_debug
Sched Debug Version: v0.11, 4.4.21-64-default #1
ktime                                   : 23533900.395978
sched_clk                               : 23543587.726648
cpu_clk                                 : 23533900.396165
jiffies                                 : 4300775771
sched_clock_stable                      : 0

sysctl_sched
  .sysctl_sched_latency                    : 6.000000
  .sysctl_sched_min_granularity            : 2.000000
  .sysctl_sched_wakeup_granularity         : 2.500000
  .sysctl_sched_child_runs_first           : 0
  .sysctl_sched_features                   : 154871
  .sysctl_sched_tunable_scaling            : 1 (logaritmic)

cpu#0, 2666.762 MHz
  .nr_running                    : 1
  .load                          : 1024
  .nr_switches                   : 1918946
[...]

cfs_rq[0]:/
  .exec_clock                    : 170176.383770
  .MIN_vruntime                  : 0.000001
  .min_vruntime                  : 347375.854324
  .max_vruntime                  : 0.000001
[...]

rt_rq[0]:/
  .rt_nr_running                 : 0
  .rt_throttled                  : 0
  .rt_time                       : 0.000000
  .rt_runtime                    : 950.000000

dl_rq[0]:
  .dl_nr_running                 : 0

  task   PID         tree-key  switches  prio     wait-time        [...]
------------------------------------------------------------------------
R  cc1 63477     98876.717832       197   120      0.000000         ...
/proc/schedstat

Displays statistics relevant to the current run queue. Also domain-specific statistics for SMP systems are displayed for all connected processors. Because the output format is not user-friendly, read the contents of /usr/src/linux/Documentation/scheduler/sched-stats.txt for more information.

/proc/PID/sched

Displays scheduling information on the process with id PID.

root # cat /proc/$(pidof gdm)/sched
gdm (744, #threads: 3)
-------------------------------------------------------------------
se.exec_start                                :          8888.758381
se.vruntime                                  :          6062.853815
se.sum_exec_runtime                          :             7.836043
se.statistics.wait_start                     :             0.000000
se.statistics.sleep_start                    :          8888.758381
se.statistics.block_start                    :             0.000000
se.statistics.sleep_max                      :          1965.987638
[...]
se.avg.decay_count                           :                 8477
policy                                       :                    0
prio                                         :                  120
clock-delta                                  :                  128
mm->numa_scan_seq                            :                    0
numa_migrations, 0
numa_faults_memory, 0, 0, 1, 0, -1
numa_faults_memory, 1, 0, 0, 0, -1

13.4 For More Information

To get a compact knowledge about Linux kernel task scheduling, you need to explore several information sources. Here are some:

  • For task scheduler System Calls description, see the relevant manual page (for example man 2 sched_setaffinity).

  • General information on scheduling is described in Scheduling wiki page.

  • A useful lecture on Linux scheduler policy and algorithm is available in http://www.inf.fu-berlin.de/lehre/SS01/OS/Lectures/Lecture08.pdf.

  • A good overview of Linux process scheduling is given in Linux Kernel Development by Robert Love (ISBN-10: 0-672-32512-8). See http://www.informit.com/articles/article.aspx?p=101760.

  • A very comprehensive overview of the Linux kernel internals is given in Understanding the Linux Kernel by Daniel P. Bovet and Marco Cesati (ISBN 978-0-596-00565-8).

  • Technical information about task scheduler is covered in files under /usr/src/linux/Documentation/scheduler.

14 Tuning the Memory Management Subsystem

To understand and tune the memory management behavior of the kernel, it is important to first have an overview of how it works and cooperates with other subsystems.

The memory management subsystem, also called the virtual memory manager, will subsequently be called VM. The role of the VM is to manage the allocation of physical memory (RAM) for the entire kernel and user programs. It is also responsible for providing a virtual memory environment for user processes (managed via POSIX APIs with Linux extensions). Finally, the VM is responsible for freeing up RAM when there is a shortage, either by trimming caches or swapping out anonymous memory.

The most important thing to understand when examining and tuning VM is how its caches are managed. The basic goal of the VM's caches is to minimize the cost of I/O as generated by swapping and file system operations (including network file systems). This is achieved by avoiding I/O completely, or by submitting I/O in better patterns.

Free memory will be used and filled up by these caches as required. The more memory is available for caches and anonymous memory, the more effectively caches and swapping will operate. However, if a memory shortage is encountered, caches will be trimmed or memory will be swapped out.

For a particular workload, the first thing that can be done to improve performance is to increase memory and reduce the frequency that memory must be trimmed or swapped. The second thing is to change the way caches are managed by changing kernel parameters.

Finally, the workload itself should be examined and tuned as well. If an application is allowed to run more processes or threads, effectiveness of VM caches can be reduced, if each process is operating in its own area of the file system. Memory overheads are also increased. If applications allocate their own buffers or caches, larger caches will mean that less memory is available for VM caches. However, more processes and threads can mean more opportunity to overlap and pipeline I/O, and may take better advantage of multiple cores. Experimentation will be required for the best results.

14.1 Memory Usage

Memory allocations in general can be characterized as pinned (also known as unreclaimable), reclaimable or swappable.

14.1.1 Anonymous Memory

Anonymous memory tends to be program heap and stack memory (for example, >malloc()). It is reclaimable, except in special cases such as mlock or if there is no available swap space. Anonymous memory must be written to swap before it can be reclaimed. Swap I/O (both swapping in and swapping out pages) tends to be less efficient than pagecache I/O, because of allocation and access patterns.

14.1.2 Pagecache

A cache of file data. When a file is read from disk or network, the contents are stored in pagecache. No disk or network access is required, if the contents are up-to-date in pagecache. tmpfs and shared memory segments count toward pagecache.

When a file is written to, the new data is stored in pagecache before being written back to a disk or the network (making it a write-back cache). When a page has new data not written back yet, it is called dirty. Pages not classified as dirty are clean. Clean pagecache pages can be reclaimed if there is a memory shortage by simply freeing them. Dirty pages must first be made clean before being reclaimed.

14.1.3 Buffercache

This is a type of pagecache for block devices (for example, /dev/sda). A file system typically uses the buffercache when accessing its on-disk metadata structures such as inode tables, allocation bitmaps, and so forth. Buffercache can be reclaimed similarly to pagecache.

14.1.4 Buffer Heads

Buffer heads are small auxiliary structures that tend to be allocated upon pagecache access. They can generally be reclaimed easily when the pagecache or buffercache pages are clean.

14.1.5 Writeback

As applications write to files, the pagecache becomes dirty and the buffercache may become dirty. When the amount of dirty memory reaches a specified number of pages in bytes (vm.dirty_background_bytes), or when the amount of dirty memory reaches a specific ratio to total memory (vm.dirty_background_ratio), or when the pages have been dirty for longer than a specified amount of time (vm.dirty_expire_centisecs), the kernel begins writeback of pages starting with files that had the pages dirtied first. The background bytes and ratios are mutually exclusive and setting one will overwrite the other. Flusher threads perform writeback in the background and allow applications to continue running. If the I/O cannot keep up with applications dirtying pagecache, and dirty data reaches a critical setting (vm.dirty_bytes or vm.dirty_ratio), then applications begin to be throttled to prevent dirty data exceeding this threshold.

14.1.6 Readahead

The VM monitors file access patterns and may attempt to perform readahead. Readahead reads pages into the pagecache from the file system that have not been requested yet. It is done to allow fewer, larger I/O requests to be submitted (more efficient). And for I/O to be pipelined (I/O performed at the same time as the application is running).

14.1.7 VFS caches

14.1.7.1 Inode Cache

This is an in-memory cache of the inode structures for each file system. These contain attributes such as the file size, permissions and ownership, and pointers to the file data.

14.1.7.2 Directory Entry Cache

This is an in-memory cache of the directory entries in the system. These contain a name (the name of a file), the inode which it refers to, and children entries. This cache is used when traversing the directory structure and accessing a file by name.

14.2 Reducing Memory Usage

14.2.1 Reducing malloc (Anonymous) Usage

Applications running on SUSE Linux Enterprise Server 12 SP5 can allocate more memory compared to SUSE Linux Enterprise Server 10. This is because of glibc changing its default behavior while allocating user space memory. See http://www.gnu.org/s/libc/manual/html_node/Malloc-Tunable-Parameters.html for explanation of these parameters.

To restore a SUSE Linux Enterprise Server 10-like behavior, M_MMAP_THRESHOLD should be set to 128*1024. This can be done with mallopt() call from the application, or via setting MALLOC_MMAP_THRESHOLD_ environment variable before running the application.

14.2.2 Reducing Kernel Memory Overheads

Kernel memory that is reclaimable (caches, described above) will be trimmed automatically during memory shortages. Most other kernel memory cannot be easily reduced but is a property of the workload given to the kernel.

Reducing the requirements of the user space workload will reduce the kernel memory usage (fewer processes, fewer open files and sockets, etc.)

14.2.3 Memory Controller (Memory Cgroups)

If the memory cgroups feature is not needed, it can be switched off by passing cgroup_disable=memory on the kernel command line, reducing memory consumption of the kernel a bit. There is also a slight performance benefit as there is a small amount of accounting overhead when memory cgroups are available even if none are configured.

14.3 Virtual Memory Manager (VM) Tunable Parameters

When tuning the VM it should be understood that some changes will take time to affect the workload and take full effect. If the workload changes throughout the day, it may behave very differently at different times. A change that increases throughput under some conditions may decrease it under other conditions.

14.3.1 Reclaim Ratios

/proc/sys/vm/swappiness

This control is used to define how aggressively the kernel swaps out anonymous memory relative to pagecache and other caches. Increasing the value increases the amount of swapping. The default value is 60.

Swap I/O tends to be much less efficient than other I/O. However, some pagecache pages will be accessed much more frequently than less used anonymous memory. The right balance should be found here.

If swap activity is observed during slowdowns, it may be worth reducing this parameter. If there is a lot of I/O activity and the amount of pagecache in the system is rather small, or if there are large dormant applications running, increasing this value might improve performance.

Note that the more data is swapped out, the longer the system will take to swap data back in when it is needed.

/proc/sys/vm/vfs_cache_pressure

This variable controls the tendency of the kernel to reclaim the memory which is used for caching of VFS caches, versus pagecache and swap. Increasing this value increases the rate at which VFS caches are reclaimed.

It is difficult to know when this should be changed, other than by experimentation. The slabtop command (part of the package procps) shows top memory objects used by the kernel. The vfs caches are the "dentry" and the "*_inode_cache" objects. If these are consuming a large amount of memory in relation to pagecache, it may be worth trying to increase pressure. Could also help to reduce swapping. The default value is 100.

/proc/sys/vm/min_free_kbytes

This controls the amount of memory that is kept free for use by special reserves including atomic allocations (those which cannot wait for reclaim). This should not normally be lowered unless the system is being very carefully tuned for memory usage (normally useful for embedded rather than server applications). If page allocation failure messages and stack traces are frequently seen in logs, min_free_kbytes could be increased until the errors disappear. There is no need for concern, if these messages are very infrequent. The default value depends on the amount of RAM.

/proc/sys/vm/watermark_scale_factor

Broadly speaking, free memory has high, low and min watermarks. When the low watermark is reached then kswapd wakes to reclaim memory in the background. It stays awake until free memory reaches the high watermark. Applications will stall and reclaim memory when the min watermark is reached.

The watermark_scale_factor defines the amount of memory left in a node/system before kswapd is woken up and how much memory needs to be free before kswapd goes back to sleep. The unit is in fractions of 10,000. The default value of 10 means the distances between watermarks are 0.1% of the available memory in the node/system. The maximum value is 1000, or 10% of memory.

Workloads that frequently stall in direct reclaim, accounted by allocstall in /proc/vmstat, may benefit from altering this parameter. Similarly, if kswapd is sleeping prematurely, as accounted for by kswapd_low_wmark_hit_quickly, then it may indicate that the number of pages kept free to avoid stalls is too low.

14.3.2 Writeback Parameters

One important change in writeback behavior since SUSE Linux Enterprise Server 10 is that modification to file-backed mmap() memory is accounted immediately as dirty memory (and subject to writeback). Whereas previously it would only be subject to writeback after it was unmapped, upon an msync() system call, or under heavy memory pressure.

Some applications do not expect mmap modifications to be subject to such writeback behavior, and performance can be reduced. Berkeley DB (and applications using it) is one known example that can cause problems. Increasing writeback ratios and times can improve this type of slowdown.

/proc/sys/vm/dirty_background_ratio

This is the percentage of the total amount of free and reclaimable memory. When the amount of dirty pagecache exceeds this percentage, writeback threads start writing back dirty memory. The default value is 10 (%).

/proc/sys/vm/dirty_background_bytes

This contains the amount of dirty memory at which the background kernel flusher threads will start writeback. dirty_background_bytes is the counterpart of dirty_background_ratio. If one of them is set, the other one will automatically be read as 0.

/proc/sys/vm/dirty_ratio

Similar percentage value as for dirty_background_ratio. When this is exceeded, applications that want to write to the pagecache are blocked and wait for kernel background flusher threads to reduce the amount of dirty memory. The default value is 20 (%).

/proc/sys/vm/dirty_bytes

This file controls the same tunable as dirty_ratio however the amount of dirty memory is in bytes as opposed to a percentage of reclaimable memory. Since both dirty_ratio and dirty_bytes control the same tunable, if one of them is set, the other one will automatically be read as 0. The minimum value allowed for dirty_bytes is two pages (in bytes); any value lower than this limit will be ignored and the old configuration will be retained.

/proc/sys/vm/dirty_expire_centisecs

Data which has been dirty in-memory for longer than this interval will be written out next time a flusher thread wakes up. Expiration is measured based on the modification time of a file's inode. Therefore, multiple dirtied pages from the same file will all be written when the interval is exceeded.

dirty_background_ratio and dirty_ratio together determine the pagecache writeback behavior. If these values are increased, more dirty memory is kept in the system for a longer time. With more dirty memory allowed in the system, the chance to improve throughput by avoiding writeback I/O and to submitting more optimal I/O patterns increases. However, more dirty memory can either harm latency when memory needs to be reclaimed or at points of data integrity (synchronization points) when it needs to be written back to disk.

14.3.3 Timing Differences of I/O Writes between SUSE Linux Enterprise 12 and SUSE Linux Enterprise 11

The system is required to limit what percentage of the system's memory contains file-backed data that needs writing to disk. This guarantees that the system can always allocate the necessary data structures to complete I/O. The maximum amount of memory that can be dirty and requires writing at any time is controlled by vm.dirty_ratio (/proc/sys/vm/dirty_ratio). The defaults are:

SLE-11-SP3:     vm.dirty_ratio = 40
SLE-12:         vm.dirty_ratio = 20

The primary advantage of using the lower ratio in SUSE Linux Enterprise 12 is that page reclamation and allocation in low memory situations completes faster as there is a higher probability that old clean pages will be quickly found and discarded. The secondary advantage is that if all data on the system must be synchronized, then the time to complete the operation on SUSE Linux Enterprise 12 will be lower than SUSE Linux Enterprise 11 SP3 by default. Most workloads will not notice this change as data is synchronized with fsync() by the application or data is not dirtied quickly enough to hit the limits.

There are exceptions and if your application is affected by this, it will manifest as an unexpected stall during writes. To prove it is affected by dirty data rate limiting then monitor /proc/PID_OF_APPLICATION/stack and it will be observed that the application spends significant time in balance_dirty_pages_ratelimited. If this is observed and it is a problem, then increase the value of vm.dirty_ratio to 40 to restore the SUSE Linux Enterprise 11 SP3 behavior.

It is important to note that the overall I/O throughput is the same regardless of the setting. The only difference is the timing of when the I/O is queued.

This is an example of using dd to asynchronously write 30% of memory to disk which would happen to be affected by the change in vm.dirty_ratio:

root # MEMTOTAL_MBYTES=`free -m | grep Mem: | awk '{print $2}'`
root # sysctl vm.dirty_ratio=40
root # dd if=/dev/zero of=zerofile ibs=1048576 count=$((MEMTOTAL_MBYTES*30/100))
2507145216 bytes (2.5 GB) copied, 8.00153 s, 313 MB/s
root # sysctl vm.dirty_ratio=20
dd if=/dev/zero of=zerofile ibs=1048576 count=$((MEMTOTAL_MBYTES*30/100))
2507145216 bytes (2.5 GB) copied, 10.1593 s, 247 MB/s

Note that the parameter affects the time it takes for the command to complete and the apparent write speed of the device. With dirty_ratio=40, more of the data is cached and written to disk in the background by the kernel. It is very important to note that the speed of I/O is identical in both cases. To demonstrate, this is the result when dd synchronizes the data before exiting:

root # sysctl vm.dirty_ratio=40
root # dd if=/dev/zero of=zerofile ibs=1048576 count=$((MEMTOTAL_MBYTES*30/100)) conv=fdatasync
2507145216 bytes (2.5 GB) copied, 21.0663 s, 119 MB/s
root # sysctl vm.dirty_ratio=20
root # dd if=/dev/zero of=zerofile ibs=1048576 count=$((MEMTOTAL_MBYTES*30/100)) conv=fdatasync
2507145216 bytes (2.5 GB) copied, 21.7286 s, 115 MB/s

Note that dirty_ratio had almost no impact here and is within the natural variability of a command. Hence, dirty_ratio does not directly impact I/O performance but it may affect the apparent performance of a workload that writes data asynchronously without synchronizing.

14.3.4 Readahead Parameters

/sys/block/<bdev>/queue/read_ahead_kb

If one or more processes are sequentially reading a file, the kernel reads some data in advance (ahead) to reduce the amount of time that processes need to wait for data to be available. The actual amount of data being read in advance is computed dynamically, based on how much "sequential" the I/O seems to be. This parameter sets the maximum amount of data that the kernel reads ahead for a single file. If you observe that large sequential reads from a file are not fast enough, you can try increasing this value. Increasing it too far may result in readahead thrashing where pagecache used for readahead is reclaimed before it can be used, or slowdowns because of a large amount of useless I/O. The default value is 512 (KB).

14.3.5 Transparent Huge Page Parameters

Transparent Huge Pages (THP) provide a way to dynamically allocate huge pages either on‑demand by the process or deferring the allocation until later via the khugepaged kernel thread. This method is distinct from the use of hugetlbfs to manually manage their allocation and use. Workloads with contiguous memory access patterns can benefit greatly from THP. A 1000-fold decrease in page faults can be observed when running synthetic workloads with contiguous memory access patterns.

There are cases when THP may be undesirable. Workloads with sparse memory access patterns can perform poorly with THP due to excessive memory usage. For example, 2 MB of memory may be used at fault time instead of 4 KB for each fault and ultimately lead to premature page reclaim. On releases older than SUSE Linux Enterprise 12 SP2, it was possible for an application to stall for long periods of time trying to allocate a THP which frequently led to a recommendation of disabling THP. Such recommendations should be re-evaluated for SUSE Linux Enterprise 12 SP3

The behavior of THP may be configured via the transparent_hugepage= kernel parameter or via sysfs. For example, it may be disabled by adding the kernel parameter transparent_hugepage=never, rebuilding your grub2 configuration, and rebooting. Verify if THP is disabled with:

root # cat /sys/kernel/mm/transparent_hugepage/enabled
always madvise [never]

If disabled, the value never is shown in square brackets like in the example above. A value of always will always try and use THP at fault time but defer to khugepaged if the allocation fails. A value of madvise will only allocate THP for address spaces explicitly specified by an application.

/sys/kernel/mm/transparent_hugepage/defrag

This parameter controls how much effort an application commits when allocating a THP. A value of always is the default for SUSE Linux Enterprise 12 SP1 and earlier releases that supported THP. If a THP is not available, the application tries to defragment memory. It potentially incurs large stalls in an application if the memory is fragmented and a THP is not available.

A value of madvise means that THP allocation requests will only defragment if the application explicitly requests it. This is the default for SUSE Linux Enterprise 12 SP2 and later releases.

defer is only available on SUSE Linux Enterprise 12 SP2 and later releases. If a THP is not available, the application will fall back to using small pages if a THP is not available. It will wake the kswapd and kcompactd kernel threads to defragment memory in the background and a THP will be allocated later by khugepaged.

The final option never will use small pages if a THP is unavailable but no other action will take place.

14.3.6 khugepaged Parameters

khugepaged will be automatically started when transparent_hugepage is set to always or madvise, and it will be automatically shut down if it is set to never. Normally this runs at low frequency but the behavior can be tuned.

/sys/kernel/mm/transparent_hugepage/khugepaged/defrag

A value of 0 will disable khugepaged even though THP may still be used at fault time. This may be important for latency-sensitive applications that benefit from THP but cannot tolerate a stall if khugepaged tries to update an application memory usage.

/sys/kernel/mm/transparent_hugepage/khugepaged/pages_to_scan

This parameter controls how many pages are scanned by khugepaged in a single pass. A scan identifies small pages that can be reallocated as THP. Increasing this value will allocate THP in the background faster at the cost of CPU usage.

/sys/kernel/mm/transparent_hugepage/khugepaged/scan_sleep_millisecs

khugepaged sleeps for a short interval specified by this parameter after each pass to limit how much CPU usage is used. Reducing this value will allocate THP in the background faster at the cost of CPU usage. A value of 0 will force continual scanning.

/sys/kernel/mm/transparent_hugepage/khugepaged/alloc_sleep_millisecs

This parameter controls how long khugepaged will sleep in the event it fails to allocate a THP in the background waiting for kswapd and kcompactd to take action.

The remaining parameters for khugepaged are rarely useful for performance tuning but are fully documented in /usr/src/linux/Documentation/vm/transhuge.txt

14.3.7 Further VM Parameters

For the complete list of the VM tunable parameters, see /usr/src/linux/Documentation/sysctl/vm.txt (available after having installed the kernel-source package).

14.4 Monitoring VM Behavior

Some simple tools that can help monitor VM behavior:

  1. vmstat: This tool gives a good overview of what the VM is doing. See Section 2.1.1, “vmstat for details.

  2. /proc/meminfo: This file gives a detailed breakdown of where memory is being used. See Section 2.4.2, “Detailed Memory Usage: /proc/meminfo for details.

  3. slabtop: This tool provides detailed information about kernel slab memory usage. buffer_head, dentry, inode_cache, ext3_inode_cache, etc. are the major caches. This command is available with the package procps.

  4. /proc/vmstat: This file gives a detailed breakdown of internal VM behaviour. The information contained within is implementation specific and may not always be available. Some information is duplicated in /proc/meminfo and other can be presented in a friendly fashion by utilities. For maximum utility, this file needs to be monitored over time to observe rates of change. The most important pieces of information that are hard to derive from other sources are as follows:

    pgscan_kswapd_*, pgsteal_kswapd_*

    These report respectively the number of pages scanned and reclaimed by kswapd since the system started. The ratio between these values can be interpreted as the reclaim efficiency with a low efficiency implying that the system is struggling to reclaim memory and may be thrashing. Light activity here is generally not something to be concerned with.

    pgscan_direct_*, pgsteal_direct_*

    These report respectively the number of pages scanned and reclaimed by an application directly. This is correlated with increases in the allocstall counter. This is more serious than kswapd activity as these events indicate that processes are stalling. Heavy activity here combined with kswapd and high rates of pgpgin, pgpout and/or high rates of pswapin or pswpout are signs that a system is thrashing heavily.

    More detailed information can be obtained using tracepoints.

    thp_fault_alloc, thp_fault_fallback

    These counters correspond to how many THPs were allocated directly by an application and how many times a THP was not available and small pages were used. Generally a high fallback rate is harmless unless the application is very sensitive to TLB pressure.

    thp_collapse_alloc, thp_collapse_alloc_failed

    These counters correspond to how many THPs were allocated by khugepaged and how many times a THP was not available and small pages were used. A high fallback rate implies that the system is fragmented and THPs are not being used even when the memory usage by applications would allow them. It is only a problem for applications that are sensitive to TLB pressure.

    compact_*_scanned, compact_stall, compact_fail, compact_success

    These counters may increase when THP is enabled and the system is fragmented. compact_stall is incremented when an application stalls allocating THP. The remaining counters account for pages scanned, the number of defragmentation events that succeeded or failed.

15 Tuning the Network

The network subsystem is complex and its tuning highly depends on the system use scenario and on external factors such as software clients or hardware components (switches, routers, or gateways) in your network. The Linux kernel aims more at reliability and low latency than low overhead and high throughput. Other settings can mean less security, but better performance.

15.1 Configurable Kernel Socket Buffers

Networking is largely based on the TCP/IP protocol and a socket interface for communication; for more information about TCP/IP, see Book “Administration Guide”, Chapter 17 “Basic Networking”. The Linux kernel handles data it receives or sends via the socket interface in socket buffers. These kernel socket buffers are tunable.

Important
Important: TCP Autotuning

Since kernel version 2.6.17 full autotuning with 4 MB maximum buffer size exists. This means that manual tuning usually will not improve networking performance considerably. It is often the best not to touch the following variables, or, at least, to check the outcome of tuning efforts carefully.

If you update from an older kernel, it is recommended to remove manual TCP tunings in favor of the autotuning feature.

The special files in the /proc file system can modify the size and behavior of kernel socket buffers; for general information about the /proc file system, see Section 2.6, “The /proc File System”. Find networking related files in:

/proc/sys/net/core
/proc/sys/net/ipv4
/proc/sys/net/ipv6

General net variables are explained in the kernel documentation (linux/Documentation/sysctl/net.txt). Special ipv4 variables are explained in linux/Documentation/networking/ip-sysctl.txt and linux/Documentation/networking/ipvs-sysctl.txt.

In the /proc file system, for example, it is possible to either set the Maximum Socket Receive Buffer and Maximum Socket Send Buffer for all protocols, or both these options for the TCP protocol only (in ipv4) and thus overriding the setting for all protocols (in core).

/proc/sys/net/ipv4/tcp_moderate_rcvbuf

If /proc/sys/net/ipv4/tcp_moderate_rcvbuf is set to 1, autotuning is active and buffer size is adjusted dynamically.

/proc/sys/net/ipv4/tcp_rmem

The three values setting the minimum, initial, and maximum size of the Memory Receive Buffer per connection. They define the actual memory usage, not only TCP window size.

/proc/sys/net/ipv4/tcp_wmem

The same as tcp_rmem, but for Memory Send Buffer per connection.

/proc/sys/net/core/rmem_max

Set to limit the maximum receive buffer size that applications can request.

/proc/sys/net/core/wmem_max

Set to limit the maximum send buffer size that applications can request.

Via /proc it is possible to disable TCP features that you do not need (all TCP features are switched on by default). For example, check the following files:

/proc/sys/net/ipv4/tcp_timestamps

TCP time stamps are defined in RFC1323.

/proc/sys/net/ipv4/tcp_window_scaling

TCP window scaling is also defined in RFC1323.

/proc/sys/net/ipv4/tcp_sack

Select acknowledgments (SACKS).

Use sysctl to read or write variables of the /proc file system. sysctl is preferable to cat (for reading) and echo (for writing), because it also reads settings from /etc/sysctl.conf and, thus, those settings survive reboots reliably. With sysctl you can read all variables and their values easily; as root use the following command to list TCP related settings:

sysctl -a | grep tcp
Note
Note: Side-Effects of Tuning Network Variables

Tuning network variables can affect other system resources such as CPU or memory use.

15.2 Detecting Network Bottlenecks and Analyzing Network Traffic

Before starting with network tuning, it is important to isolate network bottlenecks and network traffic patterns. There are some tools that can help you with detecting those bottlenecks.

The following tools can help analyzing your network traffic: netstat, tcpdump, and wireshark. Wireshark is a network traffic analyzer.

15.3 Netfilter

The Linux firewall and masquerading features are provided by the Netfilter kernel modules. This is a highly configurable rule based framework. If a rule matches a packet, Netfilter accepts or denies it or takes special action (target) as defined by rules such as address translation.

There are quite a lot of properties Netfilter can take into account. Thus, the more rules are defined, the longer packet processing may last. Also advanced connection tracking could be rather expensive and, thus, slowing down overall networking.

When the kernel queue becomes full, all new packets are dropped, causing existing connections to fail. The 'fail-open' feature allows a user to temporarily disable the packet inspection and maintain the connectivity under heavy network traffic. For reference, see https://home.regit.org/netfilter-en/using-nfqueue-and-libnetfilter_queue/.

For more information, see the home page of the Netfilter and iptables project, http://www.netfilter.org

15.4 Improving the Network Performance with Receive Packet Steering (RPS)

Modern network interface devices can move so many packets that the host can become the limiting factor for achieving maximum performance. To keep up, the system must be able to distribute the work across multiple CPU cores.

Some modern network interfaces can help distribute the work to multiple CPU cores through the implementation of multiple transmission and multiple receive queues in hardware. However, others are only equipped with a single queue and the driver must deal with all incoming packets in a single, serialized stream. To work around this issue, the operating system must "parallelize" the stream to distribute the work across multiple CPUs. On SUSE Linux Enterprise Server this is done via Receive Packet Steering (RPS). RPS can also be used in virtual environments.

RPS creates a unique hash for each data stream using IP addresses and port numbers. The use of this hash ensures that packets for the same data stream are sent to the same CPU, which helps to increase performance.

RPS is configured per network device receive queue and interface. The configuration file names match the following scheme:

/sys/class/net/<device>/queues/<rx-queue>/rps_cpus

<device> stands for the network device, such as eth0, eth1. <rx-queue> stands for the receive queue, such as rx-0, rx-1.

If the network interface hardware only supports a single receive queue, only rx-0 will exist. If it supports multiple receive queues, there will be an rx-N directory for each receive queue.

These configuration files contain a comma-delimited list of CPU bitmaps. By default, all bits are set to 0. With this setting RPS is disabled and therefore the CPU that handles the interrupt will also process the packet queue.

To enable RPS and enable specific CPUs to process packets for the receive queue of the interface, set the value of their positions in the bitmap to 1. For example, to enable CPUs 0-3 to process packets for the first receive queue for eth0, set the bit positions 0-3 to 1 in binary: 00001111. This representation then needs to be converted to hex—which results in F in this case. Set this hex value with the following command:

echo "f" > /sys/class/net/eth0/queues/rx-0/rps_cpus

If you wanted to enable CPUs 8-15:

1111 1111 0000 0000 (binary)
15     15    0    0 (decimal)
F       F    0    0 (hex)

The command to set the hex value of ff00 would be:

echo "ff00" > /sys/class/net/eth0/queues/rx-0/rps_cpus

On NUMA machines, best performance can be achieved by configuring RPS to use the CPUs on the same NUMA node as the interrupt for the interface's receive queue.

On non-NUMA machines, all CPUs can be used. If the interrupt rate is very high, excluding the CPU handling the network interface can boost performance. The CPU being used for the network interface can be determined from /proc/interrupts. For example:

root # cat /proc/interrupts
            CPU0       CPU1       CPU2       CPU3
...
  51:  113915241          0          0          0      Phys-fasteoi   eth0
...

In this case, CPU 0 is the only CPU processing interrupts for eth0, since only CPU0 contains a non-zero value.

On x86 and AMD64/Intel 64 platforms, irqbalance can be used to distribute hardware interrupts across CPUs. See man 1 irqbalance for more details.

Part VI Handling System Dumps

  • 16 Tracing Tools
  • SUSE Linux Enterprise Server comes with several tools that help you obtain useful information about your system. You can use the information for various purposes, for example, to debug and find problems in your program, to discover places causing performance drops, or to trace a running process to f…

  • 17 Kexec and Kdump
  • Kexec is a tool to boot to another kernel from the currently running one. You can perform faster system reboots without any hardware initialization. You can also prepare the system to boot to another kernel if the system crashes.

  • 18 Using systemd-coredump to Debug Application Crashes
  • systemd-coredump collects and displays kernel core dumps, for analyzing application crashes. When a process crashes (or all processes belonging to an application), its default is to log the core dump to the systemd journal, including a backtrace if possible, and to store the core dump in a file in /…

16 Tracing Tools

SUSE Linux Enterprise Server comes with several tools that help you obtain useful information about your system. You can use the information for various purposes, for example, to debug and find problems in your program, to discover places causing performance drops, or to trace a running process to find out what system resources it uses. Most of the tools are part of the installation media. In some cases, they need to be installed from the SUSE Software Development Kit, which is a separate download.

Note
Note: Tracing and Impact on Performance

While a running process is being monitored for system or library calls, the performance of the process is heavily reduced. You are advised to use tracing tools only for the time you need to collect the data.

16.1 Tracing System Calls with strace

The strace command traces system calls of a process and signals received by the process. strace can either run a new command and trace its system calls, or you can attach strace to an already running command. Each line of the command's output contains the system call name, followed by its arguments in parentheses and its return value.

To run a new command and start tracing its system calls, enter the command to be monitored as you normally do, and add strace at the beginning of the command line:

tux > strace ls
execve("/bin/ls", ["ls"], [/* 52 vars */]) = 0
brk(0)                                  = 0x618000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) \
        = 0x7f9848667000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) \
        = 0x7f9848666000
access("/etc/ld.so.preload", R_OK)      = -1 ENOENT \
(No such file or directory)
open("/etc/ld.so.cache", O_RDONLY)      = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=200411, ...}) = 0
mmap(NULL, 200411, PROT_READ, MAP_PRIVATE, 3, 0) = 0x7f9848635000
close(3)                                = 0
open("/lib64/librt.so.1", O_RDONLY)     = 3
[...]
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) \
= 0x7fd780f79000
write(1, "Desktop\nDocuments\nbin\ninst-sys\n", 31Desktop
Documents
bin
inst-sys
) = 31
close(1)                                = 0
munmap(0x7fd780f79000, 4096)            = 0
close(2)                                = 0
exit_group(0)                           = ?

To attach strace to an already running process, you need to specify the -p with the process ID (PID) of the process that you want to monitor:

tux > strace -p `pidof cron`
 Process 1261 attached
 restart_syscall(<... resuming interrupted call ...>) = 0
  stat("/etc/localtime", {st_mode=S_IFREG|0644, st_size=2309, ...}) = 0
  select(5, [4], NULL, NULL, {0, 0})      = 0 (Timeout)
  socket(PF_LOCAL, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0) = 5
  connect(5, {sa_family=AF_LOCAL, sun_path="/var/run/nscd/socket"}, 110) = 0
  sendto(5, "\2\0\0\0\0\0\0\0\5\0\0\0root\0", 17, MSG_NOSIGNAL, NULL, 0) = 17
  poll([{fd=5, events=POLLIN|POLLERR|POLLHUP}], 1, 5000) = 1 ([{fd=5, revents=POLLIN|POLLHUP}])
  read(5, "\2\0\0\0\1\0\0\0\5\0\0\0\2\0\0\0\0\0\0\0\0\0\0\0\5\0\0\0\6\0\0\0"..., 36) = 36
  read(5, "root\0x\0root\0/root\0/bin/bash\0", 28) = 28
  close(5)                                = 0
  rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
  rt_sigaction(SIGCHLD, NULL, {0x7f772b9ea890, [], SA_RESTORER|SA_RESTART, 0x7f772adf7880}, 8) = 0
  rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
  nanosleep({60, 0}, 0x7fff87d8c580)      = 0
  stat("/etc/localtime", {st_mode=S_IFREG|0644, st_size=2309, ...}) = 0
  select(5, [4], NULL, NULL, {0, 0})      = 0 (Timeout)
  socket(PF_LOCAL, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0) = 5
  connect(5, {sa_family=AF_LOCAL, sun_path="/var/run/nscd/socket"}, 110) = 0
  sendto(5, "\2\0\0\0\0\0\0\0\5\0\0\0root\0", 17, MSG_NOSIGNAL, NULL, 0) = 17
  poll([{fd=5, events=POLLIN|POLLERR|POLLHUP}], 1, 5000) = 1 ([{fd=5, revents=POLLIN|POLLHUP}])
  read(5, "\2\0\0\0\1\0\0\0\5\0\0\0\2\0\0\0\0\0\0\0\0\0\0\0\5\0\0\0\6\0\0\0"..., 36) = 36
  read(5, "root\0x\0root\0/root\0/bin/bash\0", 28) = 28
  close(5)
  [...]

The -e option understands several sub-options and arguments. For example, to trace all attempts to open or write to a particular file, use the following:

tux > strace -e trace=open,write ls ~
open("/etc/ld.so.cache", O_RDONLY)       = 3
open("/lib64/librt.so.1", O_RDONLY)      = 3
open("/lib64/libselinux.so.1", O_RDONLY) = 3
open("/lib64/libacl.so.1", O_RDONLY)     = 3
open("/lib64/libc.so.6", O_RDONLY)       = 3
open("/lib64/libpthread.so.0", O_RDONLY) = 3
[...]
open("/usr/lib/locale/cs_CZ.utf8/LC_CTYPE", O_RDONLY) = 3
open(".", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = 3
write(1, "addressbook.db.bak\nbin\ncxoffice\n"..., 311) = 311

To trace only network related system calls, use -e trace=network:

tux > strace -e trace=network -p 26520
Process 26520 attached - interrupt to quit
socket(PF_NETLINK, SOCK_RAW, 0)         = 50
bind(50, {sa_family=AF_NETLINK, pid=0, groups=00000000}, 12) = 0
getsockname(50, {sa_family=AF_NETLINK, pid=26520, groups=00000000}, \
[12]) = 0
sendto(50, "\24\0\0\0\26\0\1\3~p\315K\0\0\0\0\0\0\0\0", 20, 0,
{sa_family=AF_NETLINK, pid=0, groups=00000000}, 12) = 20
[...]

The -c calculates the time the kernel spent on each system call:

tux > strace -c find /etc -name xorg.conf
/etc/X11/xorg.conf
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 32.38    0.000181         181         1           execve
 22.00    0.000123           0       576           getdents64
 19.50    0.000109           0       917        31 open
 19.14    0.000107           0       888           close
  4.11    0.000023           2        10           mprotect
  0.00    0.000000           0         1           write
[...]
  0.00    0.000000           0         1           getrlimit
  0.00    0.000000           0         1           arch_prctl
  0.00    0.000000           0         3         1 futex
  0.00    0.000000           0         1           set_tid_address
  0.00    0.000000           0         4           fadvise64
  0.00    0.000000           0         1           set_robust_list
------ ----------- ----------- --------- --------- ----------------
100.00    0.000559                  3633        33 total

To trace all child processes of a process, use -f:

tux > strace -f rcapache2 status
execve("/usr/sbin/rcapache2", ["rcapache2", "status"], [/* 81 vars */]) = 0
brk(0)                                  = 0x69e000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) \
= 0x7f3bb553b000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) \
= 0x7f3bb553a000
[...]
[pid  4823] rt_sigprocmask(SIG_SETMASK, [],  <unfinished ...>
[pid  4822] close(4 <unfinished ...>
[pid  4823] <... rt_sigprocmask resumed> NULL, 8) = 0
[pid  4822] <... close resumed> )       = 0
[...]
[pid  4825] mprotect(0x7fc42cbbd000, 16384, PROT_READ) = 0
[pid  4825] mprotect(0x60a000, 4096, PROT_READ) = 0
[pid  4825] mprotect(0x7fc42cde4000, 4096, PROT_READ) = 0
[pid  4825] munmap(0x7fc42cda2000, 261953) = 0
[...]
[pid  4830] munmap(0x7fb1fff10000, 261953) = 0
[pid  4830] rt_sigprocmask(SIG_BLOCK, NULL, [], 8) = 0
[pid  4830] open("/dev/tty", O_RDWR|O_NONBLOCK) = 3
[pid  4830] close(3)
[...]
read(255, "\n\n# Inform the caller not only v"..., 8192) = 73
rt_sigprocmask(SIG_BLOCK, NULL, [], 8)  = 0
rt_sigprocmask(SIG_BLOCK, NULL, [], 8)  = 0
exit_group(0)

If you need to analyze the output of strace and the output messages are too long to be inspected directly in the console window, use -o. In that case, unnecessary messages, such as information about attaching and detaching processes, are suppressed. You can also suppress these messages (normally printed on the standard output) with -q. To add time stamps at the beginning of each line with a system call, use -t:

tux > strace -t -o strace_sleep.txt sleep 1; more strace_sleep.txt
08:44:06 execve("/bin/sleep", ["sleep", "1"], [/* 81 vars */]) = 0
08:44:06 brk(0)                         = 0x606000
08:44:06 mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, \
-1, 0) = 0x7f8e78cc5000
[...]
08:44:06 close(3)                       = 0
08:44:06 nanosleep({1, 0}, NULL)        = 0
08:44:07 close(1)                       = 0
08:44:07 close(2)                       = 0
08:44:07 exit_group(0)                  = ?

The behavior and output format of strace can be largely controlled. For more information, see the relevant manual page (man 1 strace).

16.2 Tracing Library Calls with ltrace

ltrace traces dynamic library calls of a process. It is used in a similar way to strace, and most of their parameters have a very similar or identical meaning. By default, ltrace uses /etc/ltrace.conf or ~/.ltrace.conf configuration files. You can, however, specify an alternative one with the -F CONFIG_FILE option.

In addition to library calls, ltrace with the -S option can trace system calls as well:

tux > ltrace -S -o ltrace_find.txt find /etc -name \
xorg.conf; more ltrace_find.txt
SYS_brk(NULL)                                              = 0x00628000
SYS_mmap(0, 4096, 3, 34, 0xffffffff)                       = 0x7f1327ea1000
SYS_mmap(0, 4096, 3, 34, 0xffffffff)                       = 0x7f1327ea0000
[...]
fnmatch("xorg.conf", "xorg.conf", 0)                       = 0
free(0x0062db80)                                           = <void>
__errno_location()                                         = 0x7f1327e5d698
__ctype_get_mb_cur_max(0x7fff25227af0, 8192, 0x62e020, -1, 0) = 6
__ctype_get_mb_cur_max(0x7fff25227af0, 18, 0x7f1327e5d6f0, 0x7fff25227af0,
0x62e031) = 6
__fprintf_chk(0x7f1327821780, 1, 0x420cf7, 0x7fff25227af0, 0x62e031
<unfinished ...>
SYS_fstat(1, 0x7fff25227230)                               = 0
SYS_mmap(0, 4096, 3, 34, 0xffffffff)                       = 0x7f1327e72000
SYS_write(1, "/etc/X11/xorg.conf\n", 19)                   = 19
[...]

You can change the type of traced events with the -e option. The following example prints library calls related to fnmatch and strlen functions:

tux > ltrace -e fnmatch,strlen find /etc -name xorg.conf
[...]
fnmatch("xorg.conf", "xorg.conf", 0)             = 0
strlen("Xresources")                             = 10
strlen("Xresources")                             = 10
strlen("Xresources")                             = 10
fnmatch("xorg.conf", "Xresources", 0)            = 1
strlen("xorg.conf.install")                      = 17
[...]

To display only the symbols included in a specific library, use -l /path/to/library:

tux > ltrace -l /lib64/librt.so.1 sleep 1
clock_gettime(1, 0x7fff4b5c34d0, 0, 0, 0)                  = 0
clock_gettime(1, 0x7fff4b5c34c0, 0xffffffffff600180, -1, 0) = 0
+++ exited (status 0) +++

You can make the output more readable by indenting each nested call by the specified number of space with the -n NUM_OF_SPACES.

16.3 Debugging and Profiling with Valgrind

Valgrind is a set of tools to debug and profile your programs so that they can run both faster and with less errors. Valgrind can detect problems related to memory management and threading, or can also serve as a framework for building new debugging tools. It is well known that this tool can incur high overhead, causing, for example, higher runtimes or changing the normal program behavior under concurrent workloads based on timing.

16.3.1 Installation

Valgrind is not shipped with standard SUSE Linux Enterprise Server distribution. To install it on your system, you need to obtain SUSE Software Development Kit, and either install it and run

zypper install VALGRIND

or browse through the SUSE Software Development Kit directory tree, locate the Valgrind package and install it with

rpm -i valgrind-VERSION_ARCHITECTURE.rpm

The SDK is a module for SUSE Linux Enterprise and is available via an online channel from the SUSE Customer Center. Alternatively download it from http://download.suse.com/. (Search for SUSE Linux Enterprise Software Development Kit). Refer to Book “Deployment Guide”, Chapter 15 “Installing Modules, Extensions, and Third Party Add-On Products” for details.

16.3.2 Supported Architectures

SUSE Linux Enterprise Server supports Valgrind on the following architectures:

  • AMD64/Intel 64

  • POWER

  • IBM Z

16.3.3 General Information

The main advantage of Valgrind is that it works with existing compiled executables. You do not need to recompile or modify your programs to use it. Run Valgrind like this:

valgrind VALGRIND_OPTIONS your-prog YOUR-PROGRAM-OPTIONS

Valgrind consists of several tools, and each provides specific functionality. Information in this section is general and valid regardless of the used tool. The most important configuration option is --tool. This option tells Valgrind which tool to run. If you omit this option, memcheck is selected by default. For example, to run find ~ -name .bashrc with Valgrind's memcheck tools, enter the following in the command line:

valgrind --tool=memcheck find ~ -name .bashrc

A list of standard Valgrind tools with a brief description follows:

memcheck

Detects memory errors. It helps you tune your programs to behave correctly.

cachegrind

Profiles cache prediction. It helps you tune your programs to run faster.

callgrind

Works in a similar way to cachegrind but also gathers additional cache-profiling information.

exp-drd

Detects thread errors. It helps you tune your multi-threaded programs to behave correctly.

helgrind

Another thread error detector. Similar to exp-drd but uses different techniques for problem analysis.

massif

A heap profiler. Heap is an area of memory used for dynamic memory allocation. This tool helps you tune your program to use less memory.

lackey

An example tool showing instrumentation basics.

16.3.4 Default Options

Valgrind can read options at start-up. There are three places which Valgrind checks:

  1. The file .valgrindrc in the home directory of the user who runs Valgrind.

  2. The environment variable $VALGRIND_OPTS

  3. The file .valgrindrc in the current directory where Valgrind is run from.

These resources are parsed exactly in this order, while later given options take precedence over earlier processed options. Options specific to a particular Valgrind tool must be prefixed with the tool name and a colon. For example, if you want cachegrind to always write profile data to the /tmp/cachegrind_PID.log, add the following line to the .valgrindrc file in your home directory:

--cachegrind:cachegrind-out-file=/tmp/cachegrind_%p.log

16.3.5 How Valgrind Works

Valgrind takes control of your executable before it starts. It reads debugging information from the executable and related shared libraries. The executable's code is redirected to the selected Valgrind tool, and the tool adds its own code to handle its debugging. Then the code is handed back to the Valgrind core and the execution continues.

For example, memcheck adds its code, which checks every memory access. As a consequence, the program runs much slower than in the native execution environment.

Valgrind simulates every instruction of your program. Therefore, it not only checks the code of your program, but also all related libraries (including the C library), libraries used for graphical environment, and so on. If you try to detect errors with Valgrind, it also detects errors in associated libraries (like C, X11, or GTK libraries). Because you probably do not need these errors, Valgrind can selectively, suppress these error messages to suppression files. The --gen-suppressions=yes tells Valgrind to report these suppressions which you can copy to a file.

You should supply a real executable (machine code) as a Valgrind argument. If your application is run, for example, from a shell or Perl script, you will by mistake get error reports related to /bin/sh (or /usr/bin/perl). In such cases, you can use --trace-children=yes to work around this issue. However, using the executable itself will avoid any confusion over this issue.

16.3.6 Messages

During its runtime, Valgrind reports messages with detailed errors and important events. The following example explains the messages:

tux > valgrind --tool=memcheck find ~ -name .bashrc
[...]
==6558== Conditional jump or move depends on uninitialised value(s)
==6558==    at 0x400AE79: _dl_relocate_object (in /lib64/ld-2.11.1.so)
==6558==    by 0x4003868: dl_main (in /lib64/ld-2.11.1.so)
[...]
==6558== Conditional jump or move depends on uninitialised value(s)
==6558==    at 0x400AE82: _dl_relocate_object (in /lib64/ld-2.11.1.so)
==6558==    by 0x4003868: dl_main (in /lib64/ld-2.11.1.so)
[...]
==6558== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)
==6558== malloc/free: in use at exit: 2,228 bytes in 8 blocks.
==6558== malloc/free: 235 allocs, 227 frees, 489,675 bytes allocated.
==6558== For counts of detected errors, rerun with: -v
==6558== searching for pointers to 8 not-freed blocks.
==6558== checked 122,584 bytes.
==6558==
==6558== LEAK SUMMARY:
==6558==    definitely lost: 0 bytes in 0 blocks.
==6558==      possibly lost: 0 bytes in 0 blocks.
==6558==    still reachable: 2,228 bytes in 8 blocks.
==6558==         suppressed: 0 bytes in 0 blocks.
==6558== Rerun with --leak-check=full to see details of leaked memory.

The ==6558== introduces Valgrind's messages and contains the process ID number (PID). You can easily distinguish Valgrind's messages from the output of the program itself, and decide which messages belong to a particular process.

To make Valgrind's messages more detailed, use -v or even -v -v.

You can make Valgrind send its messages to three different places:

  1. By default, Valgrind sends its messages to the file descriptor 2, which is the standard error output. You can tell Valgrind to send its messages to any other file descriptor with the --log-fd=FILE_DESCRIPTOR_NUMBER option.

  2. The second and probably more useful way is to send Valgrind's messages to a file with --log-file=FILENAME. This option accepts several variables, for example, %p gets replaced with the PID of the currently profiled process. This way you can send messages to different files based on their PID. %q{env_var} is replaced with the value of the related env_var environment variable.

    The following example checks for possible memory errors during the Apache Web server restart, while following children processes and writing detailed Valgrind's messages to separate files distinguished by the current process PID:

    tux > valgrind -v --tool=memcheck --trace-children=yes \
    --log-file=valgrind_pid_%p.log rcapache2 restart

    This process created 52 log files in the testing system, and took 75 seconds instead of the usual 7 seconds needed to run sudo systemctl restart apache2 without Valgrind, which is approximately 10 times more.

    tux > ls -1 valgrind_pid_*log
    valgrind_pid_11780.log
    valgrind_pid_11782.log
    valgrind_pid_11783.log
    [...]
    valgrind_pid_11860.log
    valgrind_pid_11862.log
    valgrind_pid_11863.log
  3. You may also prefer to send the Valgrind's messages over the network. You need to specify the aa.bb.cc.dd IP address and port_num port number of the network socket with the --log-socket=AA.BB.CC.DD:PORT_NUM option. If you omit the port number, 1500 will be used.

    It is useless to send Valgrind's messages to a network socket if no application is capable of receiving them on the remote machine. That is why valgrind-listener, a simple listener, is shipped together with Valgrind. It accepts connections on the specified port and copies everything it receives to the standard output.

16.3.7 Error Messages

Valgrind remembers all error messages, and if it detects a new error, the error is compared against old error messages. This way Valgrind checks for duplicate error messages. In case of a duplicate error, it is recorded but no message is shown. This mechanism prevents you from being overwhelmed by millions of duplicate errors.

The -v option will add a summary of all reports (sorted by their total count) to the end of the Valgrind's execution output. Moreover, Valgrind stops collecting errors if it detects either 1000 different errors, or 10 000 000 errors in total. If you want to suppress this limit and wish to see all error messages, use --error-limit=no.

Some errors usually cause other ones. Therefore, fix errors in the same order as they appear and re-check the program continuously.

16.4 For More Information

  • For a complete list of options related to the described tracing tools, see the corresponding man page (man 1 strace, man 1 ltrace, and man 1 valgrind).

  • To describe advanced usage of Valgrind is beyond the scope of this document. It is very well documented, see Valgrind User Manual. These pages are indispensable if you need more advanced information on Valgrind or the usage and purpose of its standard tools.

17 Kexec and Kdump

Kexec is a tool to boot to another kernel from the currently running one. You can perform faster system reboots without any hardware initialization. You can also prepare the system to boot to another kernel if the system crashes.

17.1 Introduction

With Kexec, you can replace the running kernel with another one without a hard reboot. The tool is useful for several reasons:

  • Faster system rebooting

    If you need to reboot the system frequently, Kexec can save you significant time.

  • Avoiding unreliable firmware and hardware

    Computer hardware is complex and serious problems may occur during the system start-up. You cannot always replace unreliable hardware immediately. Kexec boots the kernel to a controlled environment with the hardware already initialized. The risk of unsuccessful system start is then minimized.

  • Saving the dump of a crashed kernel

    Kexec preserves the contents of the physical memory. After the production kernel fails, the capture kernel (an additional kernel running in a reserved memory range) saves the state of the failed kernel. The saved image can help you with the subsequent analysis.

  • Booting without GRUB 2 configuration

    When the system boots a kernel with Kexec, it skips the boot loader stage. The normal booting procedure can fail because of an error in the boot loader configuration. With Kexec, you do not depend on a working boot loader configuration.

17.2 Required Packages

To use Kexec on SUSE® Linux Enterprise Server to speed up reboots or avoid potential hardware problems, make sure that the package kexec-tools is installed. It contains a script called kexec-bootloader, which reads the boot loader configuration and runs Kexec using the same kernel options as the normal boot loader.

To set up an environment that helps you obtain debug information in case of a kernel crash, make sure that the package makedumpfile is installed.

The preferred method of using Kdump in SUSE Linux Enterprise Server is through the YaST Kdump module. To use the YaST module, make sure that the package yast2-kdump is installed.

17.3 Kexec Internals

The most important component of Kexec is the /sbin/kexec command. You can load a kernel with Kexec in two different ways:

  • Load the kernel to the address space of a production kernel for a regular reboot:

    root # kexec -l KERNEL_IMAGE

    You can later boot to this kernel with kexec -e.

  • Load the kernel to a reserved area of memory:

    root # kexec -p KERNEL_IMAGE

    This kernel will be booted automatically when the system crashes.

If you want to boot another kernel and preserve the data of the production kernel when the system crashes, you need to reserve a dedicated area of the system memory. The production kernel never loads to this area because it must be always available. It is used for the capture kernel so that the memory pages of the production kernel can be preserved.

To reserve the area, append the option crashkernel to the boot command line of the production kernel. To determine the necessary values for crashkernel, follow the instructions in Section 17.4, “Calculating crashkernel Allocation Size”.

Note that this is not a parameter of the capture kernel. The capture kernel does not use Kexec.

The capture kernel is loaded to the reserved area and waits for the kernel to crash. Then, Kdump tries to invoke the capture kernel because the production kernel is no longer reliable at this stage. This means that even Kdump can fail.

To load the capture kernel, you need to include the kernel boot parameters. Usually, the initial RAM file system is used for booting. You can specify it with --initrd=FILENAME. With --append=CMDLINE, you append options to the command line of the kernel to boot.

It is helpful to include the command line of the production kernel if these options are necessary for the kernel to boot. You can simply copy the command line with --append="$(cat /proc/cmdline)" or add more options with --append="$(cat /proc/cmdline) more_options".

You can always unload the previously loaded kernel. To unload a kernel that was loaded with the -l option, use the kexec -u command. To unload a crash kernel loaded with the -p option, use kexec -p -u command.

17.4 Calculating crashkernel Allocation Size

To use Kexec with a capture kernel and to use Kdump in any way, RAM needs to be allocated for the capture kernel. The allocation size depends on the expected hardware configuration of the computer, therefore you need to specify it.

The allocation size also depends on the hardware architecture of your computer. Make sure to follow the procedure intended for your system architecture.

Procedure 17.1: Allocation Size on AMD64/Intel 64
  1. To find out the base value for the computer, run the following command:

    root # kdumptool calibrate
    Total: 49074
    Low: 72
    High: 180
    MinLow: 72
    MaxLow: 3085
    MinHigh: 0
    MaxHigh: 45824

    All values are given in megabytes.

  2. Take note of the values of Low and High.

    Note
    Note: Significance of Low and High Values

    On AMD64/Intel 64 computers, the High value stands for the memory reservation for all available memory. The Low value stands for the memory reservation in the DMA32 zone, that is, all the memory up to the 4 GB mark.

    SIZE_LOW is the amount of memory required by 32-bit-only devices. The kernel will allocate 64M for DMA32 bounce buffers. If your server does not have any 32-bit-only devices, everything should work with the default allocation of 72M for SIZE_LOW. A possible exception to this is on NUMA machines, which may make it appear that more Low memory is needed. The Kdump kernel may be booted with numa=off to make sure normal kernel allocations do not use Low memory.

  3. Adapt the High value from the previous step for the number of LUN kernel paths (paths to storage devices) attached to the computer. A sensible value in megabytes can be calculated using this formula:

    SIZE_HIGH = RECOMMENDATION + (LUNs / 2)

    The following parameters are used in this formula:

    • SIZE_HIGH.  The resulting value for High.

    • RECOMMENDATION.  The value recommended by kdumptool calibrate for High.

    • LUNs.  The maximum number of LUN kernel paths that you expect to ever create on the computer. Exclude multipath devices from this number, as these are ignored.

  4. If the drivers for your device make many reservations in the DMA32 zone, the Low value also needs to be adjusted. However, there is no simple formula to calculate these. Finding the right size can therefore be a process of trial and error.

    For the beginning, use the Low value recommended by kdumptool calibrate.

  5. The values now need to be set in the correct location.

    If you are changing the kernel command line directly

    Append the following kernel option to your boot loader configuration:

    crashkernel=SIZE_HIGH,high crashkernel=SIZE_LOW,low

    Replace the placeholders SIZE_HIGH and SIZE_LOW with the appropriate value from the previous steps and append the letter M (for megabytes).

    As an example, the following is valid:

    crashkernel=36M,high crashkernel=72M,low
    If you are using the YaST GUI:

    Set Kdump Low Memory to the determined Low value.

    Set Kdump High Memory to the determined High value.

    If you are using the YaST command line interface:

    Use the following command:

    root # yast kdump startup enable alloc_mem=LOW,HIGH

    Replace LOW with the determined Low value. Replace HIGH with the determined HIGH value.

Procedure 17.2: Allocation Size on IBM POWER and IBM Z
  1. To find out the basis value for the computer, run the following in a terminal:

    root # kdumptool calibrate

    This command returns a list of values. All values are given in megabytes.

  2. Write down the value of Low.

  3. Adapt the Low value from the previous step for the number of LUN kernel paths (paths to storage devices) attached to the computer. A sensible value in megabytes can be calculated using this formula:

    SIZE_LOW = RECOMMENDATION + (LUNs / 2)

    The following parameters are used in this formula:

    • SIZE_LOW.  The resulting value for Low.

    • RECOMMENDATION.  The value recommended by kdumptool calibrate for Low.

    • LUNs.  The maximum number of LUN kernel paths that you expect to ever create on the computer. Exclude multipath devices from this number, as these are ignored.

  4. The values now need to be set in the correct location.

    If you are working on the command line

    Append the following kernel option to your boot loader configuration:

    crashkernel=SIZE_LOW

    Replace the placeholderSIZE_LOW with the appropriate value from the previous step and append the letter M (for megabytes).

    As an example, the following is valid:

    crashkernel=108M
    If you are working in YaST

    Set Kdump Memory to the determined Low value.

Tip
Tip: Excluding Unused and Inactive CCW Devices on IBM IBM Z

Depending on the number of available devices the calculated amount of memory specified by the crashkernel kernel parameter may not be sufficient. Instead of increasing the value, you may alternatively limit the amount of devices visible to the kernel. This will lower the required amount of memory for the "crashkernel" setting.

  1. To ignore devices you can run the cio_ignore tool to generate an appropriate stanza to ignore all devices, except the ones currently active or in use.

    tux > sudo cio_ignore -u -k
    cio_ignore=all,!da5d,!f500-f502

    When you run cio_ignore -u -k, the blacklist will become active and replace any existing blacklist immediately. Unused devices are not being purged, so they still appear in the channel subsystem. But adding new channel devices (via CP ATTACH under z/VM or dynamic I/O configuration change in LPAR) will treat them as blacklisted. To prevent this, preserve the original setting by running sudo cio_ignore -l first and reverting to that state after running cio_ignore -u -k. As an alternative, add the generated stanza to the regular kernel boot parameters.

  2. Now add the cio_ignore kernel parameter with the stanza from above to KDUMP_CMDLINE_APPEND in /etc/sysconfig/kdump, for example:

    KDUMP_COMMANDLINE_APPEND="cio_ignore=all,!da5d,!f500-f502"
  3. Activate the setting by restarting kdump:

    systemctl restart kdump.service

17.5 Basic Kexec Usage

To use Kexec, ensure the respective service is enabled and running:

  • Make sure the Kexec service is loaded at system start:

    tux > sudo systemctl enable kexec-load.service
  • Make sure the Kexec service is running:

    tux > sudo systemctl start kexec-load.service

To verify if your Kexec environment works properly, try rebooting into a new Kernel with Kexec. Make sure no users are currently logged in and no important services are running on the system. If this is the case, run the following command:

systemctl kexec

The new kernel previously loaded to the address space of the older kernel rewrites it and takes control immediately. It displays the usual start-up messages. When the new kernel boots, it skips all hardware and firmware checks. Make sure no warning messages appear.

Tip
Tip: Using Kexec with the reboot Command

To make reboot use Kexec rather than performing a regular reboot, run the following command:

ln -s /usr/lib/systemd/system/kexec.target /etc/systemd/system/reboot.target

You can revert this at any time by deleting etc/systemd/system/reboot.target.

17.6 How to Configure Kexec for Routine Reboots

Kexec is often used for frequent reboots. For example, if it takes a long time to run through the hardware detection routines or if the start-up is not reliable.

Note that firmware and the boot loader are not used when the system reboots with Kexec. Any changes you make to the boot loader configuration will be ignored until the computer performs a hard reboot.

17.7 Basic Kdump Configuration

You can use Kdump to save kernel dumps. If the kernel crashes, it is useful to copy the memory image of the crashed environment to the file system. You can then debug the dump file to find the cause of the kernel crash. This is called core dump.

Kdump works similarly to Kexec (see Chapter 17, Kexec and Kdump). The capture kernel is executed after the running production kernel crashes. The difference is that Kexec replaces the production kernel with the capture kernel. With Kdump, you still have access to the memory space of the crashed production kernel. You can save the memory snapshot of the crashed kernel in the environment of the Kdump kernel.

Tip
Tip: Dumps over Network

In environments with limited local storage, you need to set up kernel dumps over the network. Kdump supports configuring the specified network interface and bringing it up via initrd. Both LAN and VLAN interfaces are supported. Specify the network interface and the mode (DHCP or static) either with YaST, or using the KDUMP_NETCONFIG option in the /etc/sysconfig/kdump file.

Important
Important: Target File System for Kdump Must Be Mounted During Configuration

When configuring Kdump, you can specify a location to which the dumped images will be saved (default: /var/crash). This location must be mounted when configuring Kdump, otherwise the configuration will fail.

17.7.1 Manual Kdump Configuration

Kdump reads its configuration from the /etc/sysconfig/kdump file. To make sure that Kdump works on your system, its default configuration is sufficient. To use Kdump with the default settings, follow these steps:

  1. Determine the amount of memory needed for Kdump by following the instructions in Section 17.4, “Calculating crashkernel Allocation Size”. Make sure to set the kernel parameter crashkernel.

  2. Reboot the computer.

  3. Enable the Kdump service:

    root # systemctl enable kdump
  4. You can edit the options in /etc/sysconfig/kdump. Reading the comments will help you understand the meaning of individual options.

  5. Execute the init script once with sudo systemctl start kdump, or reboot the system.

After configuring Kdump with the default values, check if it works as expected. Make sure that no users are currently logged in and no important services are running on your system. Then follow these steps:

  1. Switch to the rescue target with systemctl isolate rescue.target

  2. Restart the Kdump service:

    root # systemctl start kdump
  3. Unmount all the disk file systems except the root file system with:

    root # umount -a
  4. Remount the root file system in read-only mode:

    root # mount -o remount,ro /
  5. Invoke a kernel panic with the procfs interface to Magic SysRq keys:

    root # echo c > /proc/sysrq-trigger
Important
Important: Size of Kernel Dumps

The KDUMP_KEEP_OLD_DUMPS option controls the number of preserved kernel dumps (default is 5). Without compression, the size of the dump can take up to the size of the physical RAM memory. Make sure you have sufficient space on the /var partition.

The capture kernel boots and the crashed kernel memory snapshot is saved to the file system. The save path is given by the KDUMP_SAVEDIR option and it defaults to /var/crash. If KDUMP_IMMEDIATE_REBOOT is set to yes , the system automatically reboots the production kernel. Log in and check that the dump has been created under /var/crash.

17.7.1.1 Static IP Configuration for Kdump

In case Kdump is configured to use a static IP configuration from a network device, you need to add the network configuration to the KDUMP_COMMANDLINE_APPEND variable in /etc/sysconfig/kdump.

Important
Important: Changes to the Kdump Configuration File

After making changes to the /etc/sysconfig/kdump file, you need to run systemctl restart kdump.service. Otherwise, the changes will only take effect next time you reboot the system.

Example 17.1: Kdump: Example Configuration Using a Static IP Setup

The following setup has been configured:

  • eth0 has been configured with the static IP address 192.168.1.1/24

  • eth1 has been configured with the static IP address 10.50.50.100/20

  • The Kdump configuration in /etc/sysconfig/kdump looks like:

    KDUMP_CPUS=1
    KDUMP_IMMEDIATE_REBOOT=yes
    KDUMP_SAVEDIR=ftp://anonymous@10.50.50.140/crashdump/
    KDUMP_KEEP_OLD_DUMPS=5
    KDUMP_FREE_DISK_SIZE=64
    KDUMP_VERBOSE=3
    KDUMP_DUMPLEVEL=31
    KDUMP_DUMPFORMAT=lzo
    KDUMP_CONTINUE_ON_ERROR=yes
    KDUMP_NETCONFIG=eth1:static
    KDUMP_NET_TIMEOUT=30

Using this configuration, Kdump fails to reach the network when trying to write the dump to the FTP server. To solve this issue, add the network configuration to KDUMP_COMMANDLINE_APPEND in /etc/sysconfig/kdump. The general pattern for this looks like the following:

KDUMP_COMMANDLINE_APPEND='ip=CLIENT IP:SERVER IP:GATEWAY IP:NETMASK:CLIENT HOSTNAME:DEVICE:PROTOCOL'

For the example configuration this would result in:

KDUMP_COMMANDLINE_APPEND='ip=10.50.50.100:10.50.50.140:10.60.48.1:255.255.240.0:dump-client:eth1:none'

17.7.2 YaST Configuration

To configure Kdump with YaST, you need to install the yast2-kdump package. Then either start the Kernel Kdump module in the System category of YaST Control Center, or enter yast2 kdump in the command line as root.

Screenshot of the YaST Kdump Module
Figure 17.1: YaST Kdump Module: Start-Up Page

In the Start-Up window, select Enable Kdump.

The values for Kdump Memory are automatically generated the first time you open the window. However, that does not mean that they are always sufficient. To set the right values, follow the instructions in Section 17.4, “Calculating crashkernel Allocation Size”.

Important
Important: After Hardware Changes, Set Kdump Memory Values Again

If you have set up Kdump on a computer and later decide to change the amount of RAM or hard disks available to it, YaST will continue to display and use outdated memory values.

To work around this, determine the necessary memory again, as described in Section 17.4, “Calculating crashkernel Allocation Size”. Then set it manually in YaST.

Click Dump Filtering in the left pane, and check what pages to include in the dump. You do not need to include the following memory content to be able to debug kernel problems:

  • Pages filled with zero

  • Cache pages

  • User data pages

  • Free pages

In the Dump Target window, select the type of the dump target and the URL where you want to save the dump. If you selected a network protocol, such as FTP or SSH, you need to enter relevant access information as well.

Tip
Tip: Sharing the Dump Directory with Other Applications

It is possible to specify a path for saving Kdump dumps where other applications also save their dumps. When cleaning its old dump files, Kdump will safely ignore other applications' dump files.

Fill the Email Notification window information if you want Kdump to inform you about its events via e-mail and confirm your changes with OK after fine tuning Kdump in the Expert Settings window. Kdump is now configured.

17.8 Analyzing the Crash Dump

After you obtain the dump, it is time to analyze it. There are several options.

The original tool to analyze the dumps is GDB. You can even use it in the latest environments, although it has several disadvantages and limitations:

  • GDB was not specifically designed to debug kernel dumps.

  • GDB does not support ELF64 binaries on 32-bit platforms.

  • GDB does not understand other formats than ELF dumps (it cannot debug compressed dumps).

That is why the crash utility was implemented. It analyzes crash dumps and debugs the running system as well. It provides functionality specific to debugging the Linux kernel and is much more suitable for advanced debugging.

If you want to debug the Linux kernel, you need to install its debugging information package in addition. Check if the package is installed on your system with:

tux > zypper se kernel | grep debug
Important
Important: Repository for Packages with Debugging Information

If you subscribed your system for online updates, you can find debuginfo packages in the *-Debuginfo-Updates online installation repository relevant for SUSE Linux Enterprise Server 12 SP5. Use YaST to enable the repository.

To open the captured dump in crash on the machine that produced the dump, use a command like this:

crash /boot/vmlinux-4.12.14-94.37-default.gz \
/var/crash/2010-04-23-11\:17/vmcore

The first parameter represents the kernel image. The second parameter is the dump file captured by Kdump. You can find this file under /var/crash by default.

Tip
Tip: Getting Basic Information from a Kernel Crash Dump

SUSE Linux Enterprise Server ships with the utility kdumpid (included in a package with the same name) for identifying unknown kernel dumps. It can be used to extract basic information such as architecture and kernel release. It supports lkcd, diskdump, Kdump files and ELF dumps. When called with the -v switch it tries to extract additional information such as machine type, kernel banner string and kernel configuration flavor.

17.8.1 Kernel Binary Formats

The Linux kernel comes in Executable and Linkable Format (ELF). This file is usually called vmlinux and is directly generated in the compilation process. Not all boot loaders support ELF binaries, especially on the AMD64/Intel 64 architecture. The following solutions exist on different architectures supported by SUSE® Linux Enterprise Server.

17.8.1.1 AMD64/Intel 64

Kernel packages for AMD64/Intel 64 from SUSE contain two kernel files: vmlinuz and vmlinux.gz.

  • vmlinuz This is the file executed by the boot loader.

    The Linux kernel consists of two parts: the kernel itself (vmlinux) and the setup code run by the boot loader. These two parts are linked together to create vmlinuz (note the distinction: z compared to x).

    In the kernel source tree, the file is called bzImage.

  • vmlinux.gz This is a compressed ELF image that can be used by crash and GDB. The ELF image is never used by the boot loader itself on AMD64/Intel 64. Therefore, only a compressed version is shipped.

17.8.1.2 POWER

The yaboot boot loader on POWER also supports loading ELF images, but not compressed ones. In the POWER kernel package, there is an ELF Linux kernel file vmlinux. Considering crash, this is the easiest architecture.

If you decide to analyze the dump on another machine, you must check both the architecture of the computer and the files necessary for debugging.

You can analyze the dump on another computer only if it runs a Linux system of the same architecture. To check the compatibility, use the command uname -i on both computers and compare the outputs.

If you are going to analyze the dump on another computer, you also need the appropriate files from the kernel and kernel debug packages.

  1. Put the kernel dump, the kernel image from /boot, and its associated debugging info file from /usr/lib/debug/boot into a single empty directory.

  2. Additionally, copy the kernel modules from /lib/modules/$(uname -r)/kernel/ and the associated debug info files from /usr/lib/debug/lib/modules/$(uname -r)/kernel/ into a subdirectory named modules.

  3. In the directory with the dump, the kernel image, its debug info file, and the modules subdirectory, start the crash utility:

    tux > crash VMLINUX-VERSION vmcore

Regardless of the computer on which you analyze the dump, the crash utility will produce output similar to this:

tux > crash /boot/vmlinux-4.12.14-94.37-default.gz \
/var/crash/2010-04-23-11\:17/vmcore

crash 4.0-7.6
Copyright (C) 2002, 2003, 2004, 2005, 2006, 2007, 2008  Red Hat, Inc.
Copyright (C) 2004, 2005, 2006  IBM Corporation
Copyright (C) 1999-2006  Hewlett-Packard Co
Copyright (C) 2005, 2006  Fujitsu Limited
Copyright (C) 2006, 2007  VA Linux Systems Japan K.K.
Copyright (C) 2005  NEC Corporation
Copyright (C) 1999, 2002, 2007  Silicon Graphics, Inc.
Copyright (C) 1999, 2000, 2001, 2002  Mission Critical Linux, Inc.
This program is free software, covered by the GNU General Public License,
and you are welcome to change it and/or distribute copies of it under
certain conditions.  Enter "help copying" to see the conditions.
This program has absolutely no warranty.  Enter "help warranty" for details.

GNU gdb 6.1
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB.  Type "show warranty" for details.
This GDB was configured as "x86_64-unknown-linux-gnu"...

      KERNEL: /boot/vmlinux-4.12.14-94.37-default.gz
   DEBUGINFO: /usr/lib/debug/boot/vmlinux-4.12.14-94.37-default.debug
    DUMPFILE: /var/crash/2009-04-23-11:17/vmcore
        CPUS: 2
        DATE: Thu Apr 23 13:17:01 2010
      UPTIME: 00:10:41
LOAD AVERAGE: 0.01, 0.09, 0.09
       TASKS: 42
    NODENAME: eros
     RELEASE: 4.12.14-94.37-default
     VERSION: #1 SMP 2010-03-31 14:50:44 +0200
     MACHINE: x86_64  (2999 Mhz)
      MEMORY: 1 GB
       PANIC: "SysRq : Trigger a crashdump"
         PID: 9446
     COMMAND: "bash"
        TASK: ffff88003a57c3c0  [THREAD_INFO: ffff880037168000]
         CPU: 1
       STATE: TASK_RUNNING (SYSRQ)
crash> 

The command output prints first useful data: There were 42 tasks running at the moment of the kernel crash. The cause of the crash was a SysRq trigger invoked by the task with PID 9446. It was a Bash process because the echo that has been used is an internal command of the Bash shell.

The crash utility builds upon GDB and provides many additional commands. If you enter bt without any parameters, the backtrace of the task running at the moment of the crash is printed:

crash> bt
PID: 9446   TASK: ffff88003a57c3c0  CPU: 1   COMMAND: "bash"
 #0 [ffff880037169db0] crash_kexec at ffffffff80268fd6
 #1 [ffff880037169e80] __handle_sysrq at ffffffff803d50ed
 #2 [ffff880037169ec0] write_sysrq_trigger at ffffffff802f6fc5
 #3 [ffff880037169ed0] proc_reg_write at ffffffff802f068b
 #4 [ffff880037169f10] vfs_write at ffffffff802b1aba
 #5 [ffff880037169f40] sys_write at ffffffff802b1c1f
 #6 [ffff880037169f80] system_call_fastpath at ffffffff8020bfbb
    RIP: 00007fa958991f60  RSP: 00007fff61330390  RFLAGS: 00010246
    RAX: 0000000000000001  RBX: ffffffff8020bfbb  RCX: 0000000000000001
    RDX: 0000000000000002  RSI: 00007fa959284000  RDI: 0000000000000001
    RBP: 0000000000000002   R8: 00007fa9592516f0   R9: 00007fa958c209c0
    R10: 00007fa958c209c0  R11: 0000000000000246  R12: 00007fa958c1f780
    R13: 00007fa959284000  R14: 0000000000000002  R15: 00000000595569d0
    ORIG_RAX: 0000000000000001  CS: 0033  SS: 002b
crash> 

Now it is clear what happened: The internal echo command of Bash shell sent a character to /proc/sysrq-trigger. After the corresponding handler recognized this character, it invoked the crash_kexec() function. This function called panic() and Kdump saved a dump.

In addition to the basic GDB commands and the extended version of bt, the crash utility defines other commands related to the structure of the Linux kernel. These commands understand the internal data structures of the Linux kernel and present their contents in a human readable format. For example, you can list the tasks running at the moment of the crash with ps. With sym, you can list all the kernel symbols with the corresponding addresses, or inquire an individual symbol for its value. With files, you can display all the open file descriptors of a process. With kmem, you can display details about the kernel memory usage. With vm, you can inspect the virtual memory of a process, even at the level of individual page mappings. The list of useful commands is very long and many of these accept a wide range of options.

The commands that we mentioned reflect the functionality of the common Linux commands, such as ps and lsof. To find out the exact sequence of events with the debugger, you need to know how to use GDB and to have strong debugging skills. Both of these are out of the scope of this document. In addition, you need to understand the Linux kernel. Several useful reference information sources are given at the end of this document.

17.9 Advanced Kdump Configuration

The configuration for Kdump is stored in /etc/sysconfig/kdump. You can also use YaST to configure it. Kdump configuration options are available under System › Kernel Kdump in YaST Control Center. The following Kdump options may be useful for you.

You can change the directory for the kernel dumps with the KDUMP_SAVEDIR option. Keep in mind that the size of kernel dumps can be very large. Kdump will refuse to save the dump if the free disk space, subtracted by the estimated dump size, drops below the value specified by the KDUMP_FREE_DISK_SIZE option. Note that KDUMP_SAVEDIR understands the URL format PROTOCOL://SPECIFICATION, where PROTOCOL is one of file, ftp, sftp, nfs or cifs, and specification varies for each protocol. For example, to save kernel dump on an FTP server, use the following URL as a template: ftp://username:password@ftp.example.com:123/var/crash.

Kernel dumps are usually huge and contain many pages that are not necessary for analysis. With KDUMP_DUMPLEVEL option, you can omit such pages. The option understands numeric value between 0 and 31. If you specify 0, the dump size will be largest. If you specify 31, it will produce the smallest dump. For a complete table of possible values, see the manual page of kdump (man 7 kdump).

Sometimes it is very useful to make the size of the kernel dump smaller. For example, if you want to transfer the dump over the network, or if you need to save some disk space in the dump directory. This can be done with KDUMP_DUMPFORMAT set to compressed. The crash utility supports dynamic decompression of the compressed dumps.

Important
Important: Changes to the Kdump Configuration File

After making changes to the /etc/sysconfig/kdump file, you need to run systemctl restart kdump.service. Otherwise, the changes will only take effect next time you reboot the system.

17.10 For More Information

There is no single comprehensive reference to Kexec and Kdump usage. However, there are helpful resources that deal with certain aspects:

For more details on crash dump analysis and debugging tools, use the following resources:

  • In addition to the info page of GDB (info gdb), there are printable guides at https://sourceware.org/gdb/documentation/.

  • A white paper with a comprehensive description of the crash utility usage can be found at https://crash-utility.github.io/crash_whitepaper.html.

  • The crash utility also features a comprehensive online help. Use help COMMAND to display the online help for command.

  • If you have the necessary Perl skills, you can use Alicia to make the debugging easier. This Perl-based front-end to the crash utility can be found at http://alicia.sourceforge.net/ .

  • If you prefer to use Python instead, you should install Pykdump. This package helps you control GDB through Python scripts and can be downloaded from http://sf.net/projects/pykdump .

  • A very comprehensive overview of the Linux kernel internals is given in Understanding the Linux Kernel by Daniel P. Bovet and Marco Cesati (ISBN 978-0-596-00565-8).

18 Using systemd-coredump to Debug Application Crashes

systemd-coredump collects and displays kernel core dumps, for analyzing application crashes. When a process crashes (or all processes belonging to an application), its default is to log the core dump to the systemd journal, including a backtrace if possible, and to store the core dump in a file in /var/lib/systemd/coredump. You also have the option to examine the dump file with other tools such as gdb or crash (see Section 17.8, “Analyzing the Crash Dump”). There is an option to not store core dumps, but only log them in the journal, which may be useful to minimize the collection and storage of sensitive information.

18.1 Use and Configuration

systemd-coredump is enabled and ready to run by default. The default configuration is in /etc/systemd/coredump.conf:

[Coredump]
#Storage=external
#Compress=yes
#ProcessSizeMax=2G
#ExternalSizeMax=2G
#JournalSizeMax=767M
#MaxUse=
#KeepFree=

Size units are B, K, M, G, T, P, and E. ExternalSizeMax also supports a value of infinity.

The following example shows how to use Vim for simple testing, by creating a segfault to generate journal entries and a core dump.

Procedure 18.1: Creating a Coredump with Vim
  1. Enable the debuginfo-pool and debuginfo-update repositories

  2. Install vim-debuginfo and libebl1

  3. Launch vim testfile and type a few characters

  4. Get the PID and generate a segfault:

    tux > ps ax | grep vim
    2345 pts/3    S+     0:00 vim testfile
    
    root # kill -s SIGSEGV 2345
  5. List your core dumps, then examine them:

    root # coredumpctl
    TIME                        PID  UID  GID SIG PRESENT EXE
    Wed 2019-11-12 11:56:47 PST 2345 1000 100 11  *       /bin/vim
    
    root # coredumpctl info
    PID: 2345 (vim)
    UID: 0 (root)
    GID: 0 (root)
    Signal: 11 (SEGV)
    Timestamp: Wed 2019-11-12 11:58:05 PST
    Command Line: vim testfile
    Executable: /bin/vim
    Control Group: /user.slice/user-1000.slice/session-1.scope
        Unit: session-1.scope
        Slice: user-1000.slice
        Session: 1
        Owner UID: 1000 (tux)
        Boot ID: b5c251b86ab34674a2222cef102c0c88
        Machine ID: b43c44a64696799b985cafd95dc1b698
        Hostname: linux-uoch
        Coredump: /var/lib/systemd/coredump/core.vim.0.b5c251b86ab34674a2222cef102
        Message: Process 2345 (vim) of user 0 dumped core.
    
             Stack trace of thread 2345:
             #0  0x00007f21dd87e2a7 kill (libc.so.6)
             #1  0x000000000050cb35 may_core_dump (vim)
             #2  0x00007f21ddbfec70 __restore_rt (libpthread.so.0)
             #3  0x00007f21dd92ea33 __select (libc.so.6)
             #4  0x000000000050b4e3 RealWaitForChar (vim)
             #5  0x000000000050b86b mch_inchar (vim)
    [...]

When you have multiple core dumps, coredumpctl info displays all of them. Filter them by PID, COMM (command), or EXE (full path to the executable), for example, all core dumps for Vim:

root # coredumpctl info /bin/vim

See a single core dump by PID:

root # coredumpctl info 2345

Output the selected core to gdb:

root # coredumpctl gdb 2345

The asterisk in the PRESENT column indicates that a stored core dump is present. If the field is empty there is no stored core dump, and coredumpctl retrieves crash information from the journal. You may control this behavior in /etc/systemd/coredump.conf with the Storage option:

  • Storage=none, core dumps are logged in the journal, but not stored. This is useful to minimize collecting and storing sensitive information, for example for General Data Protection Regulation (GDPR) compliance.

  • Storage=external, cores are stored in /var/lib/systemd/coredump

  • Storage=journal, cores are stored in the systemd journal

  • Storage=both, cores are stored in both locations

A new instance of systemd-coredump is invoked for every core dump, so configuration changes are applied with the next core dump, and there is no need to restart any services.

Core dumps are not preserved after a system restart. You may save them permanently with coredumpctl. The following example filters by the PID and stores the core in vim.dump:

root # coredumpctl -o vim.dump dump 2345

See man systemd-coredump, man coredumpctl, and man coredump.conf for complete command and option listings.

Part VII Synchronized Clocks with Precision Time Protocol

  • 19 Precision Time Protocol
  • For network environments, it is vital to keep the computer and other devices' clocks synchronized and accurate. There are several solutions to achieve this, for example the widely used Network Time Protocol (NTP) described in Book “Administration Guide”, Chapter 26 “Time Synchronization with NTP”.

19 Precision Time Protocol

For network environments, it is vital to keep the computer and other devices' clocks synchronized and accurate. There are several solutions to achieve this, for example the widely used Network Time Protocol (NTP) described in Book “Administration Guide”, Chapter 26 “Time Synchronization with NTP”.

The Precision Time Protocol (PTP) is a protocol capable of sub-microsecond accuracy, which is better than what NTP achieves. PTP support is divided between the kernel and user space. The kernel in SUSE Linux Enterprise Server includes support for PTP clocks, which are provided by network drivers.

19.1 Introduction to PTP

The clocks managed by PTP follow a master-slave hierarchy. The slaves are synchronized to their masters. The hierarchy is updated by the best master clock (BMC) algorithm, which runs on every clock. The clock with only one port can be either master or slave. Such a clock is called an ordinary clock (OC). A clock with multiple ports can be master on one port and slave on another. Such a clock is called a boundary clock (BC). The top-level master is called the grandmaster clock. The grandmaster clock can be synchronized with a Global Positioning System (GPS). This way disparate networks can be synchronized with a high degree of accuracy.

The hardware support is the main advantage of PTP. It is supported by various network switches and network interface controllers (NIC). While it is possible to use non-PTP enabled hardware within the network, having network components between all PTP clocks PTP hardware enabled achieves the best possible accuracy.

19.1.1 PTP Linux Implementation

On SUSE Linux Enterprise Server, the implementation of PTP is provided by the linuxptp package. Install it with zypper install linuxptp. It includes the ptp4l and phc2sys programs for clock synchronization. ptp4l implements the PTP boundary clock and ordinary clock. When hardware time stamping is enabled, ptp4l synchronizes the PTP hardware clock to the master clock. With software time stamping, it synchronizes the system clock to the master clock. phc2sys is needed only with hardware time stamping to synchronize the system clock to the PTP hardware clock on the network interface card (NIC).

19.2 Using PTP

19.2.1 Network Driver and Hardware Support

PTP requires that the used kernel network driver supports either software or hardware time stamping. Moreover, the NIC must support time stamping in the physical hardware. You can verify the driver and NIC time stamping capabilities with ethtool:

ethtool -T eth0
Time stamping parameters for eth0:
Capabilities:
hardware-transmit     (SOF_TIMESTAMPING_TX_HARDWARE)
        software-transmit     (SOF_TIMESTAMPING_TX_SOFTWARE)
        hardware-receive      (SOF_TIMESTAMPING_RX_HARDWARE)
        software-receive      (SOF_TIMESTAMPING_RX_SOFTWARE)
        software-system-clock (SOF_TIMESTAMPING_SOFTWARE)
        hardware-raw-clock    (SOF_TIMESTAMPING_RAW_HARDWARE)
PTP Hardware Clock: 0
Hardware Transmit Timestamp Modes:
        off                   (HWTSTAMP_TX_OFF)
        on                    (HWTSTAMP_TX_ON)
Hardware Receive Filter Modes:
        none                  (HWTSTAMP_FILTER_NONE)
        all                   (HWTSTAMP_FILTER_ALL)

Software time stamping requires the following parameters:

SOF_TIMESTAMPING_SOFTWARE
SOF_TIMESTAMPING_TX_SOFTWARE
SOF_TIMESTAMPING_RX_SOFTWARE

Hardware time stamping requires the following parameters:

SOF_TIMESTAMPING_RAW_HARDWARE
SOF_TIMESTAMPING_TX_HARDWARE
SOF_TIMESTAMPING_RX_HARDWARE

19.2.2 Using ptp4l

ptp4l uses hardware time stamping by default. As root, you need to specify the network interface capable of hardware time stamping with the -i option. The -m tells ptp4l to print its output to the standard output instead of the system's logging facility:

ptp4l -m -i eth0
selected eth0 as PTP clock
port 1: INITIALIZING to LISTENING on INITIALIZE
port 0: INITIALIZING to LISTENING on INITIALIZE
port 1: new foreign master 00a152.fffe.0b334d-1
selected best master clock 00a152.fffe.0b334d
port 1: LISTENING to UNCALIBRATED on RS_SLAVE
master offset -25937 s0 freq +0 path delay       12340
master offset -27887 s0 freq +0 path delay       14232
master offset -38802 s0 freq +0 path delay       13847
master offset -36205 s1 freq +0 path delay       10623
master offset  -6975 s2 freq -30575 path delay   10286
port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
master offset  -4284 s2 freq -30135 path delay    9892

The master offset value represents the measured offset from the master (in nanoseconds).

The s0, s1, s2 indicators show the different states of the clock servo: s0 is unlocked, s1 is clock step, and s2 is locked. If the servo is in the locked state (s2), the clock will not be stepped (only slowly adjusted) if the pi_offset_const option is set to a negative value in the configuration file (see man 8 ptp4l for more information).

The freq value represents the frequency adjustment of the clock (in parts per billion, ppb).

The path delay value represents the estimated delay of the synchronization messages sent from the master (in nanoseconds).

Port 0 is a Unix domain socket used for local PTP management. Port 1 is the eth0 interface.

INITIALIZING, LISTENING, UNCALIBRATED and SLAVE are examples of port states which change on INITIALIZE, RS_SLAVE, and MASTER_CLOCK_SELECTED events. When the port state changes from UNCALIBRATED to SLAVE, the computer has successfully synchronized with a PTP master clock.

You can enable software time stamping with the -S option.

ptp4l -m -S -i eth3

You can also run ptp4l as a service:

systemctl start ptp4l

In this case, ptp4l reads its options from the /etc/sysconfig/ptp4l file. By default, this file tells ptp4l to read the configuration options from /etc/ptp4l.conf. For more information on ptp4l options and the configuration file settings, see man 8 ptp4l.

To enable the ptp4l service permanently, run the following:

systemctl enable ptp4l

To disable it, run

systemctl disable ptp4l

19.2.3 ptp4l Configuration File

ptp4l can read its configuration from an optional configuration file. As no configuration file is used by default, you need to specify it with -f.

ptp4l -f /etc/ptp4l.conf

The configuration file is divided into sections. The global section (indicated as [global]) sets the program options, clock options and default port options. Other sections are port specific, and they override the default port options. The name of the section is the name of the configured port—for example, [eth0]. An empty port section can be used to replace the command line option.

[global]
verbose               1
time_stamping         software
[eth0]

The example configuration file is an equivalent of the following command's options:

ptp4l -i eth0 -m -S

For a complete list of ptp4l configuration options, see man 8 ptp4l.

19.2.4 Delay Measurement

ptp4l measures time delay in two different ways: peer-to-peer (P2P) or end-to-end (E2E).

P2P

This method is specified with -P.

It reacts to changes in the network environment faster and is more accurate in measuring the delay. It is only used in networks where each port exchanges PTP messages with one other port. P2P needs to be supported by all hardware on the communication path.

E2E

This method is specified with -E. This is the default.

Automatic method selection

This method is specified with -A. The automatic option starts ptp4l in E2E mode, and changes to P2P mode if a peer delay request is received.

Important
Important: Common Measurement Method

All clocks on a single PTP communication path must use the same method to measure the time delay. A warning will be printed if either a peer delay request is received on a port using the E2E mechanism, or an E2E delay request is received on a port using the P2P mechanism.

19.2.5 PTP Management Client: pmc

You can use the pmc client to obtain more detailed information about ptp41. It reads from the standard input—or from the command line—actions specified by name and management ID. Then it sends the actions over the selected transport, and prints any received replies. There are three actions supported: GET retrieves the specified information, SET updates the specified information, and CMD (or COMMAND) initiates the specified event.

By default, the management commands are addressed to all ports. The TARGET command can be used to select a particular clock and port for the subsequent messages. For a complete list of management IDs, run pmc help.

pmc -u -b 0 'GET TIME_STATUS_NP'
sending: GET TIME_STATUS_NP
        90f2ca.fffe.20d7e9-0 seq 0 RESPONSE MANAGMENT TIME_STATUS_NP
                master_offset              283
                ingress_time               1361569379345936841
                cumulativeScaledRateOffset   +1.000000000
                scaledLastGmPhaseChange    0
                gmTimeBaseIndicator        0
                lastGmPhaseChange          0x0000'0000000000000000.0000
                gmPresent                  true
                gmIdentity                 00b058.feef.0b448a

The -b option specifies the boundary hops value in sent messages. Setting it to zero limits the boundary to the local ptp4l instance. Increasing the value will retrieve the messages also from PTP nodes that are further from the local instance. The returned information may include:

stepsRemoved

The number of communication nodes to the grandmaster clock.

offsetFromMaster, master_offset

The last measured offset of the clock from the master clock (nanoseconds).

meanPathDelay

The estimated delay of the synchronization messages sent from the master clock (nanoseconds).

gmPresent

If true, the PTP clock is synchronized to the master clock; the local clock is not the grandmaster clock.

gmIdentity

This is the grandmaster's identity.

For a complete list of pmc command line options, see man 8 pmc.

19.3 Synchronizing the Clocks with phc2sys

Use phc2sys to synchronize the system clock to the PTP hardware clock (PHC) on the network card. The system clock is considered a slave, while the network card a master. PHC itself is synchronized with ptp4l (see Section 19.2, “Using PTP”). Use -s to specify the master clock by device or network interface. Use -w to wait until ptp4l is in a synchronized state.

phc2sys -s eth0 -w

PTP operates in International Atomic Time (TAI), while the system clock uses Coordinated Universal Time (UTC). If you do not specify -w to wait for ptp4l synchronization, you can specify the offset in seconds between TAI and UTC with -O:

phc2sys -s eth0 -O -35

You can run phc2sys as a service as well:

systemctl start phc2sys

In this case, phc2sys reads its options from the /etc/sysconfig/phc2sys file. For more information on phc2sys options, see man 8 phc2sys.

To enable the phc2sys service permanently, run the following:

systemctl enable phc2sys

To disable it, run

systemctl dosable phc2sys

19.3.1 Verifying Time Synchronization

When PTP time synchronization is working properly and hardware time stamping is used, ptp4l and phc2sys output messages with time offsets and frequency adjustments periodically to the system log.

An example of the ptp4l output:

ptp4l[351.358]: selected /dev/ptp0 as PTP clock
ptp4l[352.361]: port 1: INITIALIZING to LISTENING on INITIALIZE
ptp4l[352.361]: port 0: INITIALIZING to LISTENING on INITIALIZE
ptp4l[353.210]: port 1: new foreign master 00a069.eefe.0b442d-1
ptp4l[357.214]: selected best master clock 00a069.eefe.0b662d
ptp4l[357.214]: port 1: LISTENING to UNCALIBRATED on RS_SLAVE
ptp4l[359.224]: master offset       3304 s0 freq      +0 path delay      9202
ptp4l[360.224]: master offset       3708 s1 freq  -28492 path delay      9202
ptp4l[361.224]: master offset      -3145 s2 freq  -32637 path delay      9202
ptp4l[361.224]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
ptp4l[362.223]: master offset       -145 s2 freq  -30580 path delay      9202
ptp4l[363.223]: master offset       1043 s2 freq  -28436 path delay      8972
[...]
ptp4l[371.235]: master offset        285 s2 freq  -28511 path delay      9199
ptp4l[372.235]: master offset        -78 s2 freq  -28788 path delay      9204

An example of the phc2sys output:

phc2sys[616.617]: Waiting for ptp4l...
phc2sys[628.628]: phc offset     66341 s0 freq      +0 delay   2729
phc2sys[629.628]: phc offset     64668 s1 freq  -37690 delay   2726
[...]
phc2sys[646.630]: phc offset      -333 s2 freq  -37426 delay   2747
phc2sys[646.630]: phc offset       194 s2 freq  -36999 delay   2749

ptp4l normally writes messages very frequently. You can reduce the frequency with the summary_interval directive. Its value is an exponent of the 2^N expression. For example, to reduce the output to every 1024 (which is equal to 2^10) seconds, add the following line to the /etc/ptp4l.conf file:

summary_interval 10

You can also reduce the frequency of the phc2sys command's updates with the -u SUMMARY-UPDATES option.

19.4 Examples of Configurations

This section includes several examples of ptp4l configuration. The examples are not full configuration files but rather a minimal list of changes to be made to the specific files. The string ethX stands for the actual network interface name in your setup.

Example 19.1: Slave clock using software time stamping

/etc/sysconfig/ptp4l:

OPTIONS=”-f /etc/ptp4l.conf -i ethX”

No changes made to the distribution /etc/ptp4l.conf.

Example 19.2: Slave clock using hardware time stamping

/etc/sysconfig/ptp4l:

OPTIONS=”-f /etc/ptp4l.conf -i ethX”

/etc/sysconfig/phc2sys:

OPTIONS=”-s ethX -w”

No changes made to the distribution /etc/ptp4l.conf.

Example 19.3: Master clock using hardware time stamping

/etc/sysconfig/ptp4l:

OPTIONS=”-f /etc/ptp4l.conf -i ethX”

/etc/sysconfig/phc2sys:

OPTIONS=”-s CLOCK_REALTIME -c ethX -w”

/etc/ptp4l.conf:

priority1 127
Example 19.4: Master clock using software time stamping (not generally recommended)

/etc/sysconfig/ptp4l:

OPTIONS=”-f /etc/ptp4l.conf -i ethX”

/etc/ptp4l.conf:

priority1 127

19.5 PTP and NTP

NTP and PTP time synchronization tools can coexist, synchronizing time from one to another in both directions.

19.5.1 NTP to PTP Synchronization

When ntpd is used to synchronize the local system clock, you can configure the ptp4l to be the grandmaster clock distributing the time from the local system clock via PTP. Include the priority1 option in /etc/ptp4l.conf:

[global]
priority1 127
[eth0]

Then run ptp4l:

ptp4l -f /etc/ptp4l.conf

When hardware time stamping is used, you need to synchronize the PTP hardware clock to the system clock with phc2sys:

phc2sys -c eth0 -s CLOCK_REALTIME -w

19.5.2 PTP to NTP Synchronization

You can configure ntpd to distribute the time from the system clock synchronized by ptp4l or phc2sys by using the local reference clock driver. Moreover, you need to stop ntpd from adjusting the system clock—do not specify any remote NTP servers in /etc/ntp.conf:

server   127.127.1.0
fudge    127.127.1.0 stratum 0
Note
Note: NTP and DHCP

When the DHCP client command dhclient receives a list of NTP servers, it adds them to NTP configuration by default. To prevent this behavior, set

NETCONFIG_NTP_POLICY=""

in the /etc/sysconfig/network/config file.