Jump to contentJump to page navigation: previous page [access key p]/next page [access key n]
documentation.suse.com / Optimizing Linux for AMD EPYC™ 7003 Series Processors with SUSE Linux Enterprise 15 SP2
SUSE Linux Enterprise Server 15 SP2

Optimizing Linux for AMD EPYC™ 7003 Series Processors with SUSE Linux Enterprise 15 SP2

SUSE Best Practices
Performance
Authors
Mel Gorman, Senior Kernel Engineer (SUSE)
Dario Faggioli, Software Engineer Virtualization Specialist (SUSE)
Martin Jambor, Tool Chain Developer (SUSE)
Brent Hollingsworth, Engineering Manager (AMD)
Image
SUSE Linux Enterprise 15 SP2
AMD EPYC™ 7003 Series Processors
Date: March 16, 2021
The document at hand provides an overview of the AMD EPYC* 7003 Series processors and how some computational-intensive workloads can be tuned on SUSE Linux Enterprise Server 15 SP2. Disclaimer: Documents published as part of the SUSE Best Practices series have been contributed voluntarily by SUSE employees and third parties. They are meant to serve as examples of how particular actions can be performed. They have been compiled with utmost attention to detail. However, this does not guarantee complete accuracy. SUSE cannot verify that actions described in these documents do what is claimed or whether actions described have unintended consequences. SUSE LLC, its affiliates, the authors, and the translators may not be held liable for possible errors or the consequences thereof.

1 Overview

The AMD EPYC 7003 Series processor is the latest generation of the AMD64 System-on-Chip (SoC) processor family. It is based on the Zen 3 microarchitecture introduced in 2020, supporting up to 64 cores (128 threads) and 8 memory channels per socket. At the time of writing, 1-socket and 2-socket models are available from Original Equipment Manufacturers (OEMs). This document provides an overview of the AMD EPYC 7003 Series processor and how computational-intensive workloads can be tuned on SUSE Linux Enterprise Server 15 SP2.

2 AMD EPYC 7003 Series Processor architecture

Symmetric multiprocessing (SMP) systems are those that contain two or more physical processing cores. Each core may have two threads if Symmetric Multi-Threading (SMT) is enabled with some resources being shared between SMT siblings. To minimize access latencies, multiple layers of caches are used with each level being larger but with higher access costs. Cores may share different levels of cache which should be considered when tuning for a workload.

Historically, a single socket contained several cores sharing a hierarchy of caches and memory channels and multiple sockets were connected via a memory interconnect. Modern configurations may have multiple dies as a Multi-Chip Module (MCM) with one set of interconnects within the socket and a separate interconnect for each socket. In practical terms, it means that some CPUs and memory are faster to access than others depending on the distance. This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses may not reference local memory incurring a variable access penalty.

The 3rd Generation AMD EPYC processor has an MCM design with up to nine dies on each package. From a topology point of view, this is significantly different to the 1st Generation AMD EPYC Processor design, but similar to the 2nd Generation AMD EPYC Processor. One die is a central IO die through which all off-chip communication passes through. The basic building block of a compute die is an eight-core Core CompleX (CCX) with its own L1-L3 cache hierarchy. This is double the number of cores relative to the 2nd Generation AMD EPYC Processor with the L3 cache size increasing from 16MB to 32MB to compensate. One Core Complex Die (CCD) consists of one CCX with an Infinity Link to the IO die, as opposed to two CCXs used in the 2nd Generation AMD EPYC Processor. This change in topology allowed the 3rd Generation AMD EPYC Processors to communicate directly within a CCD instead of using the IO link, thus reducing communication latency and overall effective memory access latency. A 64-core 3rd Generation AMD EPYC Processor socket therefore consists of 8 CCDs consisting of 8 CCXs or 64 cores in total (128 CPUs with SMP enabled) with one additional IO die for 9 dies in total.

Communication between the chip and memory happens via the IO die. Each CCD has one dedicated Infinity Fabric link to the die and two memory channels per CCD located on the die. The practical consequence of this architecture versus the 1st Generation AMD EPYC Processor is that the topology is simpler. The first generation had separate memory channels per die and links between dies giving two levels of NUMA distance within a single socket and a third distance when communicating between sockets. This meant that a two-socket machine for EPYC had 4 NUMA nodes (3 levels of NUMA distance). The 2nd Generation AMD EPYC Processor has only 2 NUMA nodes (2 levels of NUMA distance) which makes it easier to tune and optimize. The NUMA distances are the same for the 3rd Generation AMD EPYC Processor. The main difference to the 2nd Generation AMD Processor is that communication latencies within a CCD are reduced.

The IO die has a total of 8 memory controllers, each with 2 channels supporting one DDR4 Dual Inline Memory Modules (DIMMs) each (16 DIMMs in total). With a peak DDR4 frequency of 3.2 GHz, this yields a total of 204.8 GB/sec peak theoretical bandwidth. The exact bandwidth, however, depends on the DIMMs selected, the number of memory channels populated, and the efficiency of the application. The maximum total capacity is 4 TB.

Power management on the links is careful to minimize the amount of power required. If the links are idle, the power may be used to boost the frequency of individual cores. Hence, minimizing access is not only important from a memory access latency point of view, but it also has an impact on the speed of individual cores.

There are eight IO x16 PCIe 4.0 lanes per socket where lanes can be used as Infinity links, PCI Express links or SATA links with a limit of 32 SATA or NVMe devices. This allows very large IO configurations and a high degree of flexibility given that either IO bandwidth or the bandwidth between sockets can be optimized, depending on the OEM requirements. The most likely configuration is that the number of PCIe links will be the same for 1 and socket machines given that some lanes per socket will be used for inter-socket communication. The upshot is that a 1 or 2 socket configuration does not need to compromise on the available IO channels. The exact configuration used depends on the platform.

3 AMD EPYC 7003 Series processor topology

Figure 1, “AMD EPYC 7003 Series Processor Topology” below shows the topology of an example machine generated by the lstopo tool.

AMD EPYC 7003 Series Processor Topology
Figure 1: AMD EPYC 7003 Series Processor Topology

This tool is part of the hwloc. The two packages correspond to each socket. The CCXs consisting of 8 cores (16 threads) each should be clear, as each CCX has one L3 cache. Not obvious are the links to the IO die, but the IO die should be taken into account when splitting a workload to optimize bandwidth to memory. In this example, the IO channels are not heavily used, but the focus will be on CPU and memory-intensive loads. If optimizing for IO, it is recommended that, where possible, the workload is located on the nodes local to the IO channel.

The computer output below shows a conventional view of the topology using the numactl tool which is slightly edited for clarity. The CPU IDs that map to each node are reported on the node X cpus: lines and note the NUMA distances on the table at the bottom of the computer output. Node 0 and node 1 are a distance of 32 apart as they are on separate sockets. The distance is a not a guarantee of the access latency. But it is a rule of thumb that accesses between sockets are roughly three times the cost of accessing another die on the same socket.

epyc:~ # numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 ... 63 128 ... 191
node 0 size: 128791 MB
node 0 free: 110188 MB
node 1 cpus: 64 ... 127 192 ... 255
node 1 size: 128881 MB
node 1 free: 117322 MB
node distances:
node   0   1
  0:  10  32
  1:  32  10

Finally, the cache topology can be discovered in a variety of fashions. In addition to lstopo which can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache.

4 Memory and CPU binding

NUMA is a scalable memory architecture for multiprocessor systems that can reduce contention on a memory channel. A full discussion on tuning for NUMA is beyond the scope for this document. But the document A NUMA API for Linux at http://developer.amd.com/wordpress/media/2012/10/LibNUMA-WP-fv1.pdf provides a valuable introduction.

The default policy for programs is the local policy. A program which calls malloc() or mmap() reserves virtual address space but does not immediately allocate physical memory. The physical memory is allocated the first time the address is accessed by any thread and, if possible, the memory will be local to the accessing CPU. If the mapping is of a file, the first access may have occurred at any time in the past so there are no guarantees about locality.

When memory is allocated to a node, it is less likely to move if a thread changes to a CPU on another node or if multiple programs are remote-accessing the data unless Automatic NUMA Balancing (NUMAB) is enabled. When NUMAB is enabled, unbound process accesses are sampled. If there are enough remote accesses then the data will be migrated to local memory. This mechanism is not perfect and incurs overhead of its own. This can be important for performance for thread and process migrations between nodes to be minimized and for memory placement to be carefully considered and tuned.

The taskset tool is used to set or get the CPU affinity for new or existing processes. An example usage is to confine a new process to CPUs local to one node. Where possible, local memory will be used. But if the total required memory is larger than the node, then remote memory can still be used. In such configurations, it is recommended to size the workload such that it fits in the node. This avoids that any of the data is being paged out when kswapd wakes to reclaim memory from the local node.

numactl controls both memory and CPU policies for processes that it launches and can modify existing processes. In many respects, the parameters are easier to specify than taskset. For example, it can bind a task to all CPUs on a specified node instead of having to specify individual CPUs with taskset. Most importantly, it can set the memory allocation policy without requiring application awareness.

Using policies, a preferred node can be specified where the task will use that node if memory is available. This is typically used in combination with binding the task to CPUs on that node. If a workload's memory requirements are larger than a single node and predictable performance is required, then the interleave policy will round-robin allocations from allowed nodes. This gives suboptimal but predictable access latencies to main memory. More importantly, interleaving reduces the probability that the operating system (OS) will need to reclaim any data belonging to a large task.

Further improvements can be made to access latencies by binding a workload to a single Core Complex Die (CCD) within a node. Since L3 caches are now shared within a CCD, binding a workload to a CCD avoids L3 cache misses caused by workload migration. This is an important difference since the 2nd Generation AMD EPYC Processor which aimed to bind within a CCX.

Find examples below on how taskset and numactl can be used to start commands bound to different CPUs depending on the topology.

# Run a command bound to CPU 1
epyc:~ # taskset -c 1 [command]

# Run a command bound to CPUs belonging to node 0
epyc:~ # taskset -c `cat /sys/devices/system/node/node0/cpulist` [command]

# Run a command bound to CPUs belonging to nodes 0 and 1
epyc:~ # numactl –cpunodebind=0,1 [command]

# Run a command bound to CPUs that share L3 cache with cpu 1
epyc:~ # taskset -c `cat /sys/devices/system/cpu/cpu1/cache/index3/shared_cpu_list` [command]

4.1 Tuning for local access without binding

The ability to use local memory where possible and remote memory if necessary is valuable. But there are cases where it is imperative that local memory always be used. If this is the case, the first priority is to bind the task to that node. If that is not possible then the command sysctl vm.zone_reclaim_mode=1 can be used to aggressively reclaim memory if local memory is not available.

Note
Note: High costs

While this option is good from a locality perspective, it can incur high costs because of stalls related to reclaim and the possibility that data from the task will be reclaimed. Treat this option with a high degree of caution and testing.

4.2 Hazards with CPU binding

There are three major hazards to consider with CPU binding.

The first is to watch for remote memory nodes being used where the process is not allowed to run on CPUs local to that node. Going more in detail here is outside the scope of this paper. However, the most common scenario is an IO-bound thread communicating with a kernel IO thread on a remote node bound to the IO controller whose accesses are never local.

While tasks may be bound to CPUs, the resources they are accessing such as network or storage devices may not have interrupts routed locally. irqbalance generally makes good decisions. But in cases where network or IO is extremely high performance or the application has very low latency requirements, it may be necessary to disable irqbalance using systemctl. When that is done, the IRQs for the target device need to be routed manually to CPUs local to the target workload for optimal performance.

The second is that guides about CPU binding tend to focus on binding to a single CPU. This is not always optimal when the task communicates with other threads, as fixed bindings potentially miss an opportunity for the processes to use idle cores sharing an L1 or L2 cache. This is particularly true when dispatching IO, be it to disk or a network interface where a task may benefit from being able to migrate close to the related threads but also applies to pipeline-based communicating threads for a computational workload. Hence, focus initially on binding to CPUs sharing L3 cache. Then consider whether to bind based on a L1/L2 cache or a single CPU using the primary metric of the workload to establish whether the tuning is appropriate.

The final hazard is similar: if many tasks are bound to a smaller set of CPUs, then the subset of CPUs could be over-saturated even though the machine overall has spare capacity.

4.3 CPUsets and memory control groups

CPUsets are ideal when multiple workloads must be isolated on a machine in a predictable fashion. CPUsets allow a machine to be partitioned into subsets. These sets may overlap, and in that case they suffer from similar problems as CPU affinities. In the event there is no overlap, they can be switched to exclusive mode which treats them completely in isolation with relatively little overhead. Similarly, they are well suited when a primary workload must be protected from interference because of low-priority tasks in which case the low priority tasks can be placed in a CPUset. The caveat with CPUsets is that the overhead is higher than using scheduler and memory policies. Ordinarily, the accounting code for CPUsets is completely disabled. But when a single CPUset is created, there is a second layer of checks against scheduler and memory policies.

Similarly memcg can be used to limit the amount of memory that can be used by a set of processes. When the limits are exceeded then the memory will be reclaimed by tasks within memcg directly without interfering with any other tasks. This is ideal for ensuring there is no inference between two or more sets of tasks. Similar to CPUsets, there is some management overhead incurred. This means if tasks can simply be isolated on a NUMA boundary then this is preferred from a performance perspective. The major hazard is that, if the limits are exceeded, then the processes directly stall to reclaim the memory which can incur significant latencies.

Note
Note:

Without memcg, when memory gets low, the global reclaim daemon does work in the background and if it reclaims quickly enough, no stalls are incurred. When using memcg, observe the allocstall counter in /proc/vmstat as this can detect early if stalling is a problem.

5 High performance storage devices and interrupt affinity

High performance storage devices like Non-Volatile Memory Express (NVMe) or Serial Attached SCSI (SAS) controller are designed to take advantage of parallel IO submission. These devices typically support a large number of submit and receive queues, which are tied to MSI-X interrupts. Ideally these devices should provide as many MSI-X vectors as CPUs are in the system. To achieve the best performance each MSI-X vector should be assigned to an individual CPU.

6 Automatic NUMA balancing

Automatic NUMA Balancing will ignore any task that uses memory policies. If the workloads can be manually optimized with policies, then do so and disable automatic NUMA balancing by specifying numa_balancing=disable on the kernel command line or via sysctl. There are many cases where it is impractical or impossible to specify policies in which case the balancing should be sufficient for throughput-sensitive workloads. For latency sensitive workloads, the sampling for NUMA balancing may be too high in which case it may be necessary to disable balancing. The final corner case where NUMA balancing is a hazard is a case where the number of runnable tasks always exceeds the number of CPUs in a single node. In this case, the load balancer (and potentially affine wakes) will constantly pull tasks away from the preferred node as identified by automatic NUMA balancing resulting in excessive sampling and CPU migrations.

7 Evaluating workloads

The first and foremost step when evaluating how a workload should be tuned is to establish a primary metric such as latency, throughput or elapsed time. When each tuning step is considered or applied, it is critical that the primary metric be examined before conducting any further analysis to avoid intensive focus on the wrong bottleneck. Make sure that the metric is measured multiple times to ensure that the result is reproducible and reliable within reasonable boundaries. When that is established, analyse how the workload is using different system resources to determine what area should be the focus. The focus in this paper is on how CPU and memory is used. But other evaluations may need to consider the IO subsystem, network subsystem, system call interfaces, external libraries etc. The methodologies that can be employed to conduct this are outside the scope of the paper but the book Systems Performance: Enterprise and the Cloud by Brendan Gregg (see http://www.brendangregg.com/systems-performance-2nd-edition-book.html) is a recommended primer on the subject.

7.1 CPU utilization and saturation

Decisions on whether to bind a workload to a subset of CPUs require that the CPU utilization and any saturation risk is known. Both the ps and pidstat commands can be used to sample the number of threads in a system. Typically pidstat yields more useful information with the important exception of the run state. A system may have many threads but if they are idle then they are not contributing to utilization. The mpstat command can report the utilization of each CPU in the system.

High utilization of a small subset of CPUs may be indicative of a single-threaded workload that is pushing the CPU to the limits and may indicate a bottleneck. Conversely, low utilization may indicate a task that is not CPU-bound, is idling frequently or is migrating excessively. While each workload is different, load utilization of CPUs may show a workload that can run on a subset of CPUs to reduce latencies because of either migrations or remote accesses. When utilization is high, it is important to determine if the system could be saturated. The vmstat tool reports the number of runnable tasks waiting for CPU in the r column where any value over 1 indicates that wakeup latencies may be incurred. While the exact wakeup latency can be calculated using trace points, knowing that there are tasks queued is an important step. If a system is saturated, it may be possible to tune the workload to use fewer threads.

Overall, the initial intent should be to use CPUs from as few NUMA nodes as possible to reduce access latency but there are exceptions. The AMD EPYC 7003 Series processor has an exceptional number of high-speed memory channels to main memory. Thus consider the workload thread activity. If they are co-operating threads or sharing data, isolate them on as few nodes as possible to minimize cross-node memory accesses. If the threads are completely independent with no shared data, it may be best to isolate them on a subset of CPUs from each node. This is to maximize the number of available memory channels and throughput to main memory. For some computational workloads, it may be possible to use hybrid models such as MPI for parallelization across nodes and using openMP for threads within nodes.

7.2 Transparent Huge Pages

Huge pages are a mechanism by which performance can be improved. This is done by reducing the number of page faults, the cost of translating virtual addresses to physical addresses because of fewer layers in the page table and being able to cache translations for a larger portion of memory. Transparent Huge Pages (THP) is supported for private anonymous memory that automatically backs mappings with huge pages where anonymous memory could be allocated as heap, malloc(), mmap(MAP_ANONYMOUS), etc. While the feature has existed for a long time, it has evolved significantly.

Many tuning guides recommend disabling THP because of problems with early implementations. Specifically, when the machine was running for long enough, the use of THP could incur severe latencies and could aggressively reclaim memory in certain circumstances. These problems have been resolved by the time SUSE Linux Enterprise Server 15 SP2 was released. This means there are no good grounds for automatically disabling THP because of severe latency issues without measuring the impact. However, there are exceptions that are worth considering for specific workloads.

Some high-end in-memory databases and other applications aggressively use mprotect() to ensure that unprivileged data is never leaked. If these protections are at the base page granularity then there may be many THP splits and rebuilds that incur overhead. It can be identified if this is a potential problem by using strace to detect the frequency and granularity of the system call. If they are high frequency then consider disabling THP. It can also be sometimes inferred from observing the thp_split and thp_collapse_alloc counters in /proc/vmstat.

Workloads that sparsely address large mappings may have a higher memory footprint when using THP. This could result in premature reclaim or fallback to remote nodes. An example would be HPC workloads operating on large sparse matrices. If memory usage is much higher than expected then compare memory usage with and without THP to decide if the trade off is not worthwhile. This may be critical on the AMD EPYC 7003 Series processor given that any spillover will congest the Infinity links and potentially cause cores to run at a lower frequency.

Note
Note: Sparsely addressed memory

This is specific to sparsely addressed memory. A secondary hint for this case may be that the application primarily uses large mappings with a much higher Virtual Size (VSZ, see Section 7.4, “Memory utilization and saturation”) than Resident Set Size (RSS). Applications which densely address memory benefit from the use of THP by achieving greater bandwidth to memory.

Parallelized workloads that operate on shared buffers with threads using more CPUs that are on a single node may experience a slowdown with THP if the granularity of partitioning is not aligned to the huge page. The problem is that if a large shared buffer is partitioned on a 4K boundary then false sharing may occur whereby one thread accesses a huge page locally and other threads access it remotely. If this situation is encountered, it is preferable that the granularity of sharing is increased to the THP size. But if that is not possible then disabling THP is an option.

Applications that are extremely latency sensitive or must always perform in a deterministic fashion can be hindered by THP. While there are fewer faults, the time for each fault is higher as memory must be allocated and cleared before being visible. The increase in fault times may be in the microsecond granularity. Ensure this is a relevant problem as it typically only applies to extremely latency-sensitive applications. The secondary problem is that a kernel daemon periodically scans a process looking for contiguous regions that can be backed by huge pages. When creating a huge page, there is a window during which that memory cannot be accessed by the application and new mappings cannot be created until the operation is complete. This can be identified as a problem with thread-intensive applications that frequently allocate memory. In this case consider effectively disabling khugepaged by setting a large value in /sys/kernel/mm/transparent_hugepage/khugepaged/alloc_sleep_millisecs. This will still allow THP to be used opportunistically while avoiding stalls when calling malloc() or mmap().

THP can be disabled. To do so, specify transparent_hugepage=disable on the kernel command line, at runtime via /sys/kernel/mm/transparent_hugepage/enabled or on a per process basis by using a wrapper to execute the workload that calls prctl(PR_SET_THP_DISABLE).

7.3 User/kernel footprint

Assuming an application is mostly CPU or memory bound, it is useful to determine if the footprint is primarily in user space or kernel space. This gives a hint where tuning should be focused. The percentage of CPU time can be measured on a coarse-grained fashion using vmstat or a fine-grained fashion using mpstat. If an application is mostly spending time in user space then the focus should be on tuning the application itself. If the application is spending time in the kernel then it should be determined which subsystem dominates. The strace or perf trace commands can measure the type, frequency and duration of system calls as they are the primary reasons an application spends time within the kernel. In some cases, an application may be tuned or modified to reduce the frequency and duration of system calls. In other cases, a profile is required to identify which portions of the kernel are most relevant as a target for tuning.

7.4 Memory utilization and saturation

The traditional means of measuring memory utilization of a workload is to examine the Virtual Size (VSZ) and Resident Set Size (RSS) using either the ps or pidstat tool. This is a reasonable first step but is potentially misleading when shared memory is used and multiple processes are examined. VSZ is simply a measure of memory space reservation and is not necessarily used. RSS may be double accounted if it is a shared segment between multiple processes. The file /proc/pid/maps can be used to identify all segments used and whether they are private or shared. The file /proc/pid/smaps yields more detailed information including the Proportional Set Size (PSS). PSS is an estimate of RSS except it is divided between the number of processes mapping that segment which can give a more accurate estimate of utilization. Note that the smaps file is very expensive to read and should not be monitored at a high frequency. Finally, the Working Set Size (WSS) is the amount of memory active required to complete computations during an arbitrary phase of a programs execution. It is not a value that can be trivially measured. But conceptually it is useful as the interaction between WSS relative to available memory affects memory residency and page fault rates.

On NUMA systems, the first saturation point is a node overflow when the local policy is in effect. Given no binding of memory, when a node is filled, a remote node’s memory will be used transparently and background reclaim will take place on the local node. Two consequences of this are that remote access penalties will be used and old memory from the local node will be reclaimed. If the WSS of the application exceeds the size of a local node then paging and re-faults may be incurred.

The first thing to identify is that a remote node overflow occurred which is accounted for in /proc/vmstat as the numa_hit, numa_miss, numa_foreign, numa_interleave, numa_local and numa_other counters:

  • numa_hit is incremented when an allocation uses the preferred node where preferred may be either a local node or one specified by a memory policy.

  • numa_miss is incremented when an alternative node is used to satisfy an allocation.

  • numa_foreign is rarely useful but is accounted against a node that was preferred. It is a subtle distinction from numa_miss that is rarely useful.

  • numa_interleave is incremented when an interleave policy was used to select allowed nodes in a round-robin fashion.

  • numa_local increments when a local node is used for an allocation regardless of policy.

  • numa_other is used when a remote node is used for an allocation regardless of policy.

For the local memory policy, the numa_hit and numa_miss counters are the most important to pay attention to. An application that is allocating memory that starts incrementing the numa_miss implies that the first level of saturation has been reached. If monitoring the proc is undesirable then the numastat provides the same information. If this is observed on the AMD EPYC 7003 Series Processor, it may be valuable to bind the application to nodes that represent dies on a single socket. If the ratio of hits to misses is close to 1, consider an evaluation of the interleave policy to avoid unnecessary reclaim.

Note
Note: NUMA statistics

These NUMA statistics only apply at the time a physical page is allocated and is not related to the reference behaviour of the workload. For example, if a task running on node 0 allocates memory local to node 0 then it will be accounted for as a node_hit in the statistics. However, if the memory is shared with a task running on node 1, all the accesses may be remote, which is a miss from the perspective of the hardware but not accounted for in /proc/vmstat. Detecting remote and local accesses at a hardware level requires using the hardware's Performance Management Unit to detect.

When the first saturation point is reached then reclaim will be active. This can be observed by monitoring the pgscan_kswapd and pgsteal_kswapd /proc/vmstat counters. If this is matched with an increase in major faults or minor faults then it may be indicative of severe thrashing. In this case the interleave policy should be considered. An ideal tuning option is to identify if shared memory is the source of the usage. If this is the case, then interleave the shared memory segments. This can be done in some circumstances using numactl or by modifying the application directly.

More severe saturation is observed if the pgscan_direct and pgsteal_direct counters are also increasing as these indicate that the application is stalling while memory is being reclaimed. If the application was bound to individual nodes, increasing the number of available nodes will alleviate the pressure. If the application is unbound, it indicates that the WSS of the workload exceeds all available memory. It can only be alleviated by tuning the application to use less memory or increasing the amount of RAM available.

As before, whether to use memory nodes from one socket or two sockets depends on the application. If the individual processes are independent then either socket can be used. But where possible, keep communicating processes on the same socket to maximize memory throughput while minimizing the socket interconnect traffic.

7.5 Other resources

The analysis of other resources is outside the scope of this paper. However, a common scenario is that an application is IO-bound. A superficial check can be made using the vmstat tool. This tool checks what percentage of CPU time is spent idle combined with the number of processes that are blocked and the values in the bi and bo columns. Further analysis is required to determine if an application is IO rather than CPU or memory bound. But this is a sufficient check to start with.

8 Power management

Modern CPUs balance power consumption and performance through Performance States (P-States). Low utilization workloads may use lower P-States to conserve power while still achieving acceptable performance. When a CPU is idle, lower power idle states (C-States) can be selected to further conserve power. However this comes with higher exit latencies when lower power states are selected. It is further complicated by the fact that, if individual cores are idle and running at low power, the additional power can be used to boost the performance of active cores. This means this scenario is not a straight-forward balance between power consumption and performance. More complexity is added on the AMD EPYC 7003 Series processor whereby spare power may be used to boost either cores or the Infinity links.

The 3rd Generation AMD EPYC processor provides SenseMI which, among other capabilities, enables CPUs to make adjustments to voltage and frequency depending on the historical state of the CPU. There is a latency penalty when switching P-States, but the AMD EPYC 7003 Series processor is capable of fine-grained in the adjustments that can be made to reduce likelihood that the latency is a bottleneck. On SUSE Linux Enterprise Server, the AMD EPYC 7003 Series processor uses the acpi_cpufreq driver. This allows P-states to be configured to match requested performance. However, this is limited in terms of the full capabilities of the hardware. It cannot boost the frequency beyond the maximum stated frequencies, and if a target is specified, then the highest frequency below the target will be used. A special case is if the governor is set to performance. In this situation the hardware will use the highest available frequency in an attempt to work quickly and then return to idle.

What should be determined is whether power management is likely to be a factor for a workload. One that is limited to a subset of active CPUs and nodes will have high enough utilization. This means that power management will not be active on those cores and no action is required. Hence, with CPU binding, the issue of power management may be side-stepped.

Secondly, a workload that does not communicate heavily with other processes and is mostly CPU-bound will also not experience any side effects because of power management.

The workloads that are most likely to be affected are those that synchronously communicate between multiple threads or those that idle frequently and have low CPU utilization overall. It will be further compounded if the threads are sensitive to wakeup latency. However, there are secondary effects if a workload must complete quickly but the CPU is running at a low frequency.

The P-State and C-State of each CPU can be examined using the turbostat utility. The computer output below shows an example, slightly edited to fit the page, where a workload is busy on CPU 0 and other workloads are idle. A useful exercise is to start a workload and monitor the output of turbostat paying close attention to CPUs that have moderate utilization and running at a lower frequency. If the workload is latency-sensitive, it is grounds for either minimizing the number of CPUs available to the workload or configuring power management.

Pac. Die  Core CPU Avg_M Busy% Bzy_M TSC_M IRQ   POLL C1   C2    POLL% C1%   C2%
-    -    -    -   37    1.86  1989  1996  27650 11   9989 60041 0.00   0.34 99.22
0    0    0    0   34    1.35  2534  1996  97    0    0    205   0.00   0.00 99.75
0    0    0    128 31    1.25  2494  1996  46    0    0    181   0.00   0.00 99.90
0    0    1    1   3630 100.00 3630  1996  1747  0    0    0     0.00   0.00 0.00
0    0    1    129 86    2.73  3146  1996  21    0    0    172   0.00   0.00 99.95
0    0    2    2   29    1.18  2475  1996  23    0    0    171   0.00   0.00 99.95
0    0    2    130 29    1.19  2476  1996  25    0    0    172   0.00   0.00 99.95
0    0    3    3   30    1.21  2476  1996  23    0    0    171   0.00   0.00 99.96
0    0    3    131 30    1.21  2476  1996  24    0    0    172   0.00   0.00 99.96

In the event it is determined that tuning CPU frequency management is appropriate. Then the following actions can be taken to set the management policy to performance using the cpupower utility:

epyc:~# cpupower frequency-set -g performance
Setting cpu: 0
Setting cpu: 1
Setting cpu: 2
...

Persisting it across reboots can be done via a local init script, via udev or via one-shot systemd service file if it is deemed to be necessary. Note that turbostat will still show that idling CPUs use a low frequency. The impact of the policy is that the highest P-State will be used as soon as possible when the CPU is active. In some cases, a latency bottleneck will occur because of a CPU exiting idle. If this is identified on the AMD EPYC 7003 Series processor, restrict the C-state by specifying processor.max_cstate=2 on the kernel command line. This will prevent CPUs from entering lower C-states. It is expected on the AMD EPYC 7003 Series Processor that the exit latency from C1 is very low. But by allowing C2, it reduces interference from the idle loop injecting micro-operations into the pipeline and should be the best state overall. It is also possible to set the max idle state on individual cores using cpupower idle-set. If SMT is enabled, the idle state should be set on both siblings.

9 Security mitigation

On occasion, a security fix is applied to a distribution that has a performance impact. The most notable example is Meltdown and multiple variants of Spectre but includes others such as ForeShadow (L1TF). The AMD EPYC 7003 Series processor is immune to the Meltdown variant. Page table isolation is never active. However, it is vulnerable to Spectre variants. The following table lists all security vulnerabilities that affect the 3rd Generation AMD EPYC processor and which mitgations are enabled by default for SUSE Linux Enterprise Server 15 SP2.

Table 1: Security mitigations for AMD EPYC 7003 Series Processors
VulnerabilityAffected Mitigations

ITLB Multihit

No

N/A

L1TF

No

N/A

MDS

No

N/A

Meltdown

No

N/A

Speculative Store Bypass

Yes

prctl and seccomp policy

Spectre v1

Yes

usercopy/swapgs barriers and __user pointer sanitization

Spectre v2

Yes

retpoline, RSB filling, and conditional IBPB, IBRS_FW, and STIBP

SRBDS

No

N/A

TSX Async Abort

No

N/A

If it can be guaranteed that the server is in a trusted environment running only known code that is not malicious, the mitigations=off parameter can be specified on the kernel command line. This option disables all security mitigations and might improve performance, but the gains for the AMD EPYC 7003 Series processor will only be marginal when compared against other CPUs.

10 Hardware-based profiling

The AMD EPYC 7003 Series processor has extensive Performance Monitoring Unit (PMU) capabilities, and advanced monitoring of a workload can be conducted via the perf. The command supports a range of hardware events including cycles, L1 cache access/misses, TLB access/misses, retired branch instructions and mispredicted branches. To identify what subsystem may be worth tuning in the OS, the most useful invocation is perf record -a -e cycles sleep 30 to capture 30 seconds of data for the entire system. You can also call perf record -e cycles command to gather a profile of a given workload. Specific information on the OS can be gathered through tracepoints or creating probe points with perf or trace-cmd. But the details on how to conduct such analysis are beyond the scope of this paper.

11 Compiler selection

SUSE Linux Enterprise ships with multiple versions of GCC. SUSE Linux Enterprise 15 services packs ship with GCC 7 which at the time of writing is GCC 7.5.0 with the package version 7-3.3.22. The intention is to avoid unintended consequences when porting code that may affect the building and operation of applications. The GCC 7 development originally started in 2016, with a branch created in 2017 and GCC 7.5 released in 2019. This means that the system compiler has no awareness of the AMD EPYC 7002 or 7003 Series processor.

Fortunately, the add-on Developer Tools Module includes additional compilers with the latest version currently based on GCC 10.2. This compiler is capable of generating optimised code targeted at the 2nd Generation AMD EPYC processor and provides additional support for OpenMP 5.0. Unlike the system compiler, the major version of GCC shipped with the Developer Tools Module can change during the lifetime of the product. It is expected that GCC 11 will be included in future releases for generating optimised code for the 3rd Generation AMD EPYC processor.

The OS packages are built against a generic target but where applications and benchmarks can be rebuilt from source, the minimum option should be -march=znver2 for GCC 10 or -march=znver3 for GCC 11 and later versions of GCC.

Further information on how to install the Developer Tools Module and how to build optimised versions of applications can be found on the guide Advanced optimization and new capabilities of GCC 10.

12 Candidate workloads

The workloads that will benefit most from the 3rd Generation AMD EPYC processor architecture are those that can be parallelized and are either memory or IO-bound. This is particularly true for workloads that are NUMA friendly. They can be parallelized, and each thread can operate independently for the majority of the workloads lifetime. For memory-bound workloads, the primary benefit will be taking advantage of the high bandwidth available on each channel. For IO-bound workloads, the primary benefit will be realised when there are multiple storage devices, each of which is connected to the node local to a task issuing IO.

12.1 Test setup

The following sections will demonstrate how an OpenMP and MPI workload can be configured and tuned on an AMD EPYC 7003 Series processor reference platform. The system has two AMD EPYC 7713 processors, each with 64 cores and SMT enabled for a total of 128 cores (256 logical CPUs). Half the memory banks are populated with DDR4 2666 MHz RDIMMs, with a theoretical maximum transfer speed of 21.3 GB/sec each. With single ranking, the peak memory transfer speed is 341 GB/sec. Note that the peak theoretical transfer speed is rarely reached in practice given that it can be affected by the mix of reads/writes and the location and temporal proximity of memory locations accessed.

Table 2: Test setup

CPU

2x AMD EPYC 7713

Platform

AMD Engineering Sample Platform

Drive

Micron 1100 SSD

OS

SUSE Linux Enterprise Server 15 SP2

Memory Interleaving

Channel

Memory Speed

2667MHz (single rank)

Kernel command line

mitigations=off

12.2 Test workload: STREAM

STREAM is a memory bandwidth benchmark created by Dr. John D. McCalpin from the University of Virginia (for more information, see https://www.cs.virginia.edu/stream/. It can be used to measure bandwidth of each cache level and bandwidth to main memory while calculating four basic vector operations. Each operation can exhibit different throughputs to main memory depending on the locations and type of access.

The benchmark was configured to run both single-threaded and parallelized with OpenMP to take advantage of multiple memory controllers. The array elements for the benchmark was set at 268,435,456 elements at compile time so each that array was 2048MB in size for a total memory footprint of approximately 6144 MB. The size was selected in line with the recommendation from STREAM that the array sizes be at least 4 times the total size of L3 cache available in the system. An array-size offset was used so that the separate arrays for each parallelized thread would not share a Transparent Huge Page. The reason for that is that NUMA balancing may choose to migrate shared pages leading to some distortion of the results.

Table 3: Test workload: STREAM

Compiler

gcc (SUSE Linux Enterprise) 10.2.1

Compiler flags

-Ofast -march=znver2 -mcmodel=medium -DOFFSET=512

OpenMP compiler flag

-fopemp

OpenMP environment variables

OMP_PROC_BIND=SPREAD

OMP_NUM_THREADS=32

The march=znver2 is a reflection of the compiler available in SUSE LInux Enterprise 15 SP2 at the time of writing but it should be checked if GCC 11 in the Developer Tools Module. The number of openMP threads was selected to have at least one thread running for every memory channel by having one thread per L3 cache available. The OMP_PROC_BIND parameter spreads the threads such that one thread is bound to each available dedicated L3 cache to maximize available bandwidth. This can be verified using trace-cmd, as illustrated below with slight editing for formatting and clarity.

epyc:~ # trace-cmd record -e sched:sched_migrate_task ./stream
epyc:~ # trace-cmd report
...
stream-nnn x: sched_migrate_task: comm=stream pid=137530 prio=120 orig_cpu=0 dest_cpu=8
stream-nnn x: sched_migrate_task: comm=stream pid=137531 prio=120 orig_cpu=0 dest_cpu=16
stream-nnn x: sched_migrate_task: comm=stream pid=137532 prio=120 orig_cpu=0 dest_cpu=24
stream-nnn x: sched_migrate_task: comm=stream pid=137533 prio=120 orig_cpu=0 dest_cpu=32
stream-nnn x: sched_migrate_task: comm=stream pid=137534 prio=120 orig_cpu=0 dest_cpu=40
stream-nnn x: sched_migrate_task: comm=stream pid=137535 prio=120 orig_cpu=0 dest_cpu=48
stream-nnn x: sched_migrate_task: comm=stream pid=137536 prio=120 orig_cpu=0 dest_cpu=56
stream-nnn x: sched_migrate_task: comm=stream pid=137537 prio=120 orig_cpu=0 dest_cpu=64
stream-nnn x: sched_migrate_task: comm=stream pid=137538 prio=120 orig_cpu=0 dest_cpu=72
stream-nnn x: sched_migrate_task: comm=stream pid=137539 prio=120 orig_cpu=0 dest_cpu=80
stream-nnn x: sched_migrate_task: comm=stream pid=137540 prio=120 orig_cpu=0 dest_cpu=88
stream-nnn x: sched_migrate_task: comm=stream pid=137541 prio=120 orig_cpu=0 dest_cpu=96
stream-nnn x: sched_migrate_task: comm=stream pid=137542 prio=120 orig_cpu=0 dest_cpu=104
stream-nnn x: sched_migrate_task: comm=stream pid=137543 prio=120 orig_cpu=0 dest_cpu=112
stream-nnn x: sched_migrate_task: comm=stream pid=137544 prio=120 orig_cpu=0 dest_cpu=120
...

Several options were considered for the test system that were unnecessary for STREAM running on the AMD EPYC 7003 Series processor but may be useful in other situations. STREAM performance can be limited if a load/store instruction stalls to fetch data. CPUs may automatically prefetch data based on historical behaviour but it is not guaranteed. In limited cases depending on the CPU and workload, this may be addressed by specifying -fprefetch-loop-arrays and depending on whether the workload is store-intensive, -mprefetchwt1. In the test system using AMD EPYC 7003 Series processors, explicit prefetching did not help and was omitted. This is because an explicitly scheduled prefetch may disable a CPUs predictive algorithms and degrade performance. Similarly, for some workloads branch mispredictions can be a major problem, and in some cases breach mispredictions can be offset against I-Cache pressure by specifying -funroll-loops. In the case of STREAM on the test system, the CPU accurately predicted the branches rendering the unrolling of loops unnecessary. For math-intensive workloads it can be beneficial to link the application with -lmvec depending on the application. In the case of STREAM, the workload did not use significant math-based operations and so this option was not used. Some styles of code blocks and loops can also be optimized to use vectored operations by specifying -ftree-vectorize and explicitly adding support for CPU features such as -mavx2. In all cases, STREAM does not benefit as its operations are very basic. But it should be considered on an application-by-application basis and when building support libraries such as numerical libraries. In all cases, experimentation is recommended but caution advised, particularly when considering options like prefetch that may have been advisable on much older CPUs or completely different workloads. Such options are not universally beneficial or always suitable for modern CPUs such as the AMD EPYC 7003 Series processors.

In the case of STREAM running on the AMD EPYC 7003 Series processor, it was sufficient to enable -Ofast. This includes the -O3 optimizations to enable vectorization. In addition, it gives some leeway for optimizations that increase the code size with additional options for fast math that may not be standards-compliant.

For OpenMP, the SPREAD option was used to spread the load across L3 caches. Spreading across memory controllers may have been more optimal but that information is not easily discoverable. OpenMP has a variety of different placement options if manually tuning placement. But careful attention should be paid to OMP_PLACES given the importance of the L3 Cache topology in AMD EPYC 7003 Series Processors. At the time of writing, it is not possible to specify l3cache as a place similar to what MPI has. In this case, the topology will need to be examined either with library support such as hwloc, directly via the sysfs or manually. While it is possible to guess via the CPUID, it is not recommended as CPUs may be offlined or the enumeration may change. An example specification of places based on L3 cache for the test system is;

{0:8,128:8},  {8:8,136:8},   {16:8,144:8},  {24:8,152:8}, {32:8,160:8}, {40:8,168:8},
{48:8,176:8}, {56:8,184:8},  {64:8,192:8},  {72:8,200:8}, {80:8,208:8}, {88:8,216:8},
{96:8,224:8}, {104:8,232:8}, {112:8,240:8}, {120:8,248:8}

Figure 2, “STREAM Bandwidth, Single Threaded and Parallelized” shows the reported bandwidth for the single and parallelized case. The single-threaded bandwidth for the basic Copy vector operation on a single core was 39.5 GB/sec. This is higher than the theoretical maximum of a single DIMM, but the IO die may interleave accesses. Channels are interleaved in this configuration. This has been recommended as the best balance for a variety of workloads but limits the absolute maximum of a specialized benchmark like STREAM. The total throughput for each parallel operation ranged from 173 GB/sec to 263 GB/sec depending on the type of operation and how efficiently memory bandwidth was used.

Note
Note: STREAM scores

Higher STREAM scores can be reported by reducing the array sizes so that cache is partially used with the maximum score requiring that each threads memory footprint fits inside the L1 cache. Additionally, it is possible to achieve results closer to the theoretical maximum by manual optimization of the STREAM benchmark using vectored instructions and explicit scheduling of loads and stores. The purpose of this configuration was to illustrate the impact of properly binding a workload that can be fully parallelized with data-independent threads.

STREAM Bandwidth, Single Threaded and Parallelized
Figure 2: STREAM Bandwidth, Single Threaded and Parallelized

12.3 Test workload: NASA Parallel Benchmark

NASA Parallel Benchmark (NPB) is a small set of compute-intensive kernels designed to evaluate the performance of supercomputers. They are small compute kernels derived from Computational Fluid Dynamics (CFD) applications. The problem size can be adjusted for different memory sizes. Reference implementations exist for both MPI and OpenMP. This setup will focus on the MPI reference implementation.

While each application behaves differently, one common characteristic is that the workload is very context-switch intensive, barriers frequently yield the CPU to other tasks and the lifetime of individual processes can be very short-lived. The following paragraphs detail the tuning selected for this workload.

The most basic step is setting the CPU governor to performance although it is not mandatory. This can address issues with short-lived, or mobile tasks, failing to run long enough for a higher P-State to be selected even though the workload is very throughput sensitive. The migration cost parameter is set to reduce the frequency the load balancer will move an individual task. The minimum granularity is adjusted to reduce over-scheduling effects.

Depending on the computational kernel used, the workload may require a power-of-two number or a square number of processes to be used. However, note that using all available CPUs can mean that the application can contend with itself for CPU time. Furthermore, as IO may be issued to shared memory backed by disk, there are system threads that also need CPU time. Finally, if there is CPU contention, MPI tasks can be stalled waiting on an available CPU and OpenMPI may yield tasks prematurely if it detects there are more MPI tasks than CPUs available. These factors should be carefully considered when tuning for parallelized workloads in general and MPI workloads in particular.

In the specific case of testing NPB on the System Under Test, there was usually limited advantage to limiting the number of CPUs used. For the Embarrassingly Parallel (EP) load in particular, it benefits from using all available CPUs. Hence, the default configuration used all available CPUs (256) which is both a power-of-two and square number of CPUs because it was a sensible starting point. However, this is not universally true. Using perf, it was found that some workloads are memory-bound and do not benefit from a high degree of parallelization. In addition, for the final configuration, some workloads were parallelized to have one task per L3 cache in the system to maximize cache usage. The exception was the Scalar Pentadiagonal (SP) workload which was both memory bound and benefited from using all available CPUs. Also, note that the System Under Test used a pre-production CPU, that all memory channels were not populated and the memory speed was not the maximum speed supported by the platform. Hence, the optimal choices for this test may not be universally true for all AMD EPYC platforms. This highlights that there is no universal good choice for optimizing a workload for a platform and that experimentation and validation of tuning parameters is vital.

The basic compilation flags simply turned on all safe optimizations. The tuned flags used -Ofast which can be unsafe for some mathematical workloads but generated the correct output for NPB. The other flags used the optimal instructions available on the distribution compiler and vectorized some operations. GCC 10 is more strict in terms of matching types in Fortran. Depending on the version of NPB used, it may be necessary to specify the -fallow-argument-mismatch or -fallow-invalid-boz to compile unless the source code is modified.

As NPB uses shared files, an XFS partition was used for the temporary files. It is, however, only used for mapping shared files and is not a critical path for the benchmark and no IO tuning is necessary. In some cases with MPI applications, it will be possible to use a tmpfs partition for OpenMPI. This avoids unnecessary IO assuming the increased physical memory usage does not cause the application to be paged out.

Table 4: Test workload: NASA Parallel Benchmark

Compiler

gcc (SUSE Linux Enterprise) 10.2.1

OpenMPI

openmpi2-2.1.6-10.19.x86_64

Default compiler flags

-m64 -O3 -mcmodel=large

Default number processes

256

Selective number processes

bt=256 sp=256 mg=32 lu=121 ep=256 cg=32

Fortran compiler flags

-fallow-argument-mismatch -fallow-invalid-boz

Tuned compiler flags

-Ofast -march=znver2 -mtune=znver2 -ftree-vectorize

CPU governor performance

cpupower frequency-set -g performance

Scheduler parameters

sysctl -w kernel.sched_migration_cost_ns=5000000

sysctl -w kernel.sched_min_granularity_ns=10000000

mpirun parameters

-mca btl ^openib,udapl -np 64 --bind-to l3cache:overload-allowed

mpirun environment

TMPDIR=/xfs-data-partition

Figure 3, “NAS MPI Results” shows the time, as reported by the benchmark, for each of the kernels to complete.

NAS MPI Results
Figure 3: NAS MPI Results

The baseline test used all available CPUs, basic compilation options and the performance governor. The second test tuned used tuned compilation options, scheduler tuning and bound tasks to L3 caches gaining between 0.9% and 17.54% performance on average. The final test selective used processes that either used all CPUs, avoided heavy overloaded or limited processes to one per L3 cache showing overall gains between 0.9% and 21.6% depending on the computational kernel. Despite the binding, automatic NUMA balancing was not tuned as the majority of accesses were local to the NUMA domain and the overhead of NUMA balancing was negligible.

In some cases, it will be necessary to compile an application that can run on different CPUs. In such cases, -march=znver2 may not be suitable if it generates binaries that are incompatible with other vendors. In such cases, it is possible to specify the ISA-specific options that are cross-compatible with many x86-based CPUs such as -mavx2, -mfma, -msse2 or msse4a while favouring optimal code generation for AMD EPYC 7003 Series processors with -mtune=znver2. This can be used to strike a balance between excellent performance on a single CPU and great performance on multiple CPUs.

13 Using AMD EPYC 7003 Series processors for virtualization

Running Virtual Machines (VMs) has some aspects in common with running regular tasks on a host operating system. Therefore, most of the tuning advice described so far in this document are valid and applicable to this section too.

However, virtualization workloads do pose their own specific challenges and some special considerations need to be made, to achieve a better tailored and more effective set of tuning advice, for cases where a server is used only as a virtualization host. And this is especially relevant for large systems, such as AMD EPYC 7003 Series processors.

This is because:

  • VMs typically run longer, and consume more memory, than most of others regular OS processes.

  • VMs can be configured to behave either as NUMA-aware or non NUMA-aware workloads.

In fact, VMs often run for hours, days, or even months, without being terminated or restarted. Therefore, it is almost never acceptable to pay the price of suboptimal resource partitioning and allocation, even when there is the expectation that things will be better next time. For example, it is always desirable that vCPUs run close to the memory that they are accessing. For reasonably big NUMA aware VMs, this happens only with proper mapping of the virtual NUMA nodes of the guest to physical NUMA nodes on the host. For smaller, NUMA-unaware VMs, that means allocating all their memory on the smallest possible number of host NUMA nodes, and making their vCPUs run on the pCPUs of those nodes as much as possible.

Also, poor mapping of virtual machine resources (virtual CPUs and memory, but also IO) on the host topology induces performance degradation to everything that runs inside the virtual machine – and potentially even to other components of the system .

Regarding NUMA-awareness, a VM is called out to be NUMA aware if a (virtual) NUMA topology is defined and exposed to the VM itself and if the OS that the VM runs (guest OS) is also NUMA aware. On the contrary, a VM is called NUMA-unaware if either no (virtual) NUMA topology is exposed or the guest OS is not NUMA aware.

VMs that are large enough (in terms of amount of memory and number of virtual CPUs) to span multiple host NUMA nodes, benefit from being configured as NUMA aware VMs. However, even for small and NUMA-unaware VMs, wise placement of their memory on the host nodes, and effective mapping of their virtual CPUs (vCPUs) on the host physical CPUs (pCPUs) is key for achieving good and consistent performance.

This second half of the present document focuses on tuning a system where CPU and memory intensive workloads run inside VMs. We will explain how to configure and tune both the host and the VMs, in a way that performance comparable to the ones of the host can be achieved.

Both the Kernel-based Virtual Machine (KVM) and the Xen-Project hypervisor (Xen), as available in SUSE Linux Enterprise Server 15 SP2, provide mechanisms to enact this kind of resource partitioning and allocation.

Note
Note: Virtual Machine types

KVM only supports one type of VM – a fully hardware-based virtual machine (HVM). Under Xen, VMs can be paravirtualized (PV) or hardware virtualized machines (HVM).

Xen HVM guests with paravirtualized interfaces enabled (called PVHVM, or HVM) are very similar to KVM VMs, which are based on hardware virtualization but also employ paravirtualized IO (VirtIO). In this document, we always refer to Xen VMs of the (PV)HVM type.

14 Resources allocation and tuning of the host

No details are given, here, about how to install and configure a system so that it becomes a suitable virtualization host. For similar instructions, refer to the SUSE documentation at SUSE Linux Enterprise Server 15 SP2 Virtualization Guide: Installation of Virtualization Components.

The same applies to configuring things such as networking and storage, either for the host or for the VMs. For similar instructions, refer to suitable chapters of OS and virtualization documentation and manuals. As an example, to know how to assign network interfaces (or ports) to one or more VMs, for improved network throughput, refer to SUSE Linux Enterprise Server 15 SP2 Virtualization Guide: Assigning a Host PCI Device to a VM Guest.

14.1 Allocating resources to the host OS

Even if the main purpose of a server is limited to running VMs, some activities will be carried out on the host OS. In fact, in both Xen and KVM, the host OS is at least responsible for helping with the IO of the VMs. It may, therefore, be necessary to make sure that the host OS has some resources (namely, CPUs and memory) exclusively assigned to itself.

Note
Note: Host OS on KVM and on Xen

On KVM, the host OS is the SUSE Linux Enterprise distribution installed on the server, which then loads the hypervisor kernel modules. On Xen, the host OS still is a SUSE Linux Enterprise distribution, but it runs inside what is to all the effect an (although special) Virtual Machine (called Domain 0, or Dom0).

In the absence of any specific requirements involving host resources, a good rule of thumb suggests that 5 to 10 percent of the physical RAM should be reserved to the host OS. On KVM, increase that quota in case the plan is to run many (for example hundreds) of VMs. On Xen, it is okay to always give dom0 not more than a few GigaBytes of memory. This is especially the case when planning to take advantage of disaggregation (see https://wiki.xenproject.org/wiki/Dom0_Disaggregation).

In terms of CPUs, depending on the workload, it may be fine to use all the physical CPUs for the VMs (and this is in fact how the benchmarks in the experimental section of this guide have been conducted). On the other hand, if it is necessary to keep some CPUs for the host Os / dom0, we advise to reserve at least one physical core on each NUMA node. If that is not possible, consider reserving at least one core on the NUMA node(s) that have IO channels attached to it (them). In fact, host OS activities are mostly related to performing IO for VMs. It is beneficial for performance if the kernel threads that handle devices can run on the node to which the devices themselves are attached. As an example, in the system shown in Figure 1, “AMD EPYC 7003 Series Processor Topology”, this would mean reserving at least one physical core for the host OS on node 0.

System administrators need to be able to reach out and login to the system, to monitor, manage and troubleshoot it. Therefore, it is possible that even more resources needs to be assigned to the host OS. This would be for making sure that management tools (for example, the SSH daemon) can be reached, and that the hypervisor's toolstack (for example, libvirt) can run without too much contention.

14.1.1 Reserving CPUs and memory for the host on KVM

When using KVM, sparing, for example, 2 cores and 12 GB of RAM for the host OS is done by stopping creating VMs when the total number of vCPUs of all VMs has reached 252 (as each core has 2 threads) or when the total cumulative amount of allocated RAM has reached 220 GB.

To make sure that these CPUs are not used by the virtual machines and are available to the host, virtual CPU pinning (discussed later in this document) can be used. There are also other ways to enforce this, for example with cgroups, or by means of the isolcpus boot parameter, but we will not cover them in details.

14.1.2 Reserving CPUs and memory for the host on Xen

When using Xen, dom0 resource allocation needs to be done explicitly at system boot time. For example, assigning 8 physical cores and 12 GB of RAM to it is done by specifying the following additional parameters on the hypervisor boot command line (for example, by editing /etc/defaults/grub, and then updating the boot loader):

dom0_mem=12288M,max:12288M dom0_max_vcpus=16

The number of vCPUs is 16 because we want Dom0 to have 8 physical cores, and the AMD EPYC 7003 Series Processor has Symmetric MultiThreading (SMT). 12288M memory (that is 12 GB) is specified both as current and as maximum value, to prevent Dom0 from using ballooning (see https://wiki.xenproject.org/wiki/Tuning_Xen_for_Performance#Memory ).

Making sure that Dom0 vCPUs run on specific pCPUs is not strictly necessary. If wanted, it can be enforced acting on the Xen scheduler, when Dom0 is booted (as there is currently no mechanism to set up this via Xen boot time parameters). If using the xl toolstack, the command is:

xl vcpu-pin 0 <vcpu-ID> <pcpu-list>

Or, with virsh:

virsh vcpupin 0 --vcpu <vcpu-ID> --cpulist <pcpu-list>

Note that virsh vcpupin –config ... is not effective for Dom0.

If wanting to limit Dom0 to only a specific (set of) NUMA node(s), the dom0_nodes=<nodeid> boot command line option can be used. This will affect both memory and vCPUs. In fact, it means that memory for Dom0 will be allocated on the specified node(s), and the vCPUs of Dom0 will be restricted to run on those same node(s). When the Dom0 has booted, it is still possible to use xl vcpu-pin or virsh vcpupin to change where its vCPUs will be scheduled. But the memory will never be moved from where it has been allocated during boot. On AMD EPYC 7003 Series processors, using this option is not recommended.

14.1.3 Reserving CPUs for the host under IO intensive VM workloads

Covering the cases where VMs run IO intensive workloads, is out of the scope of this guide. As general advice, if IO done in VM is important, it may be appropriate to leave to the host OS either one core or one thread for each IO device used by each VM. If this is not possible, for example because it reduces to much the CPUs that remain available for running VMs (especially in case the goal is to run many of them), then exploring alternative solutions for handling IO devices (such as SR-IOV) is recommended.

14.2 (Transparent) Huge Pages

For virtualization workloads, rather than using Transparent Huge Pages (THP) on the host, it is recommended that 1 GB huge pages are used for the memory of the VMs. This sensibly reduces both the page table management overhead and the level of resource contention that the system faces when VMs update their page tables. Moreover, if the host is entirely devoted to running VMs, Huge Pages are likely not required for host OS processes. Actually, in this case, having them active on the host may even negatively affect performance, as the THP service daemon (which is there for merging regular pages and form Huge Pages) can interfere with VMs' vCPUs execution. For disabling Huge Pages for the host OS, in a KVM setup, add the following host kernel command-line option:

transparent_hugepage=never

Or execute this command, at runtime:

echo never > /sys/kernel/mm/transparent_hugepage/enabled

To use 1 GB Huge Pages as backing memory of KVM guests, such pages need to be allocated at the host OS level. It is best to make that happen by adding the following boot time parameter:

default_hugepagesz=1GB hugepagesz=1GB hugepages=<number of hugepages>

The value <number of hugepages> can be computed by taking the amount of memory we want to devoted to VMs, and dividing it by the page size (1 GB). For example on our host with 256 GB of RAM, creating 200 Huge Pages means we can accommodate up to 210 GB of VMs' RAM, and leave plenty (~50GB) to the host OS.

On Xen, none of the above is necessary. In fact, Dom0 is a paravirtualized guest, for which Huge Pages support is not present. On the other hand, for the memory used for the hypervisor and for the memory allocated by the hypervisor for HVM VMs, Huge Pages are always used as much as possible. So no explicit tuning is needed.

14.3 Automatic NUMA balancing

On Xen, Automatic NUMA Balancing (NUMAB) in the host OS should be disabled. Dom0 is, currently, a paravirtualized guest without a (virtual) NUMA topology and, thus, NUMAB would be totally useless.

On KVM, NUMAB can be useful in dynamic virtualization scenarios, where VMs are created, destroyed and re-created relatively quickly. Or, in general, it can be useful in cases where it is not possible or desirable to statically partition the host resources and assign them explicitly to VMs. This, however, comes at the price of some latency being introduced. Plus, NUMAB's own operation can interfere with VMs' execution and cause further performance degradation. In any case, this document focuses on achieving the best possible performance for VMs, through specific tuning and static resource allocation, therefore it is recommended to leave NUMAB turned off. This can be done by adding the following parameter to the host kernel command line:

numa_balancing=disable

If anything changes and the system is repurposed to achieve different goals, NUMAB can be enabled on-line with the command:

echo 0 > /proc/sys/kernel/numa_balancing

14.4 Services, daemons and power management

Service daemons have been discussed already, in the first part of the document. Most of the consideration done there, applies here as well.

For example, tuned should either not be used, or the profile should be set to one that does not implicitly put the CPUs in polling mode. Both throughput-performance and virtual-host profiles from SUSE Linux Enterprise Server 15 SP2 are okay, as neither of them touches /dev/cpu_dma_latency.

irqbalance can be a source of latency, for no significant performance improvement. The suggestion is again to disable it for latency sensitive workloads. This means, though, that IRQs may need to be manually bound to the appropriate CPUs, considering the IO topology.

As far as power management is concerned, the cpufreq governor can either be kept as it is by default, or switched to performance, depending on the nature of the workloads of interest.

Note
Note: Power management

For anything that concerns power management on KVM, changing the tuned profile or using cpupower, from the host OS will have the effects one can expect, and described already in the first half of the document. On Xen, CPU frequency scaling is enacted by the hypervisor. It can be controlled from within Dom0, but the tool that needs to be used is different. For example, for setting governor to performance, we need the following:

xenpm set-scaling-governor performance

14.5 Secure Encrypted Virtualization (SEV)

Secure Encrypted Virtualization (SEV) technology, supported on the 3rd Generation AMD EPYC Processors, allows the memory of the VMs to be encrypted. Enabling it for a VM happens in the VM's own configuration file, but there are preparation steps that needs to occur at the host level.

The libvirt documentation contains all the necessary steps required for enabling SEV on a host that supports it. It has been enough to add the following boot parameters to the host kernel command line, in the boot loader:

mem_encrypt=on kvm_amd.sev=1

For further details, refer to Libvirt documentation: Enabling SEV on the host.

It should be noted that, at least as far as the workload analyzed in this document, enabling SEV on the host has not caused any noticeable performance degradation. In fact, running CPU and memory intensive workloads, both on the host and in VMs, with or without the parameters above added to the host kernel boot command line, resulted in indistinguishable results. Therefore, enabling SEV at the host level can be considered safe, from the point of view of not hurting performance of VMs and workloads that will not enable VM memory encryption. Of course, this applies to when SEV is enabled for the host and is not used for encrypting VMs.

What happens when SEV is used for encrypting VMs' memory will be described later in the guide.

Note
Note: SEV on Xen

In SUSE Linux Enterprise Server 15 SP2, SEV is not available on Xen.

15 Resources allocation and tuning of VMs

For instructions on how to create an initial VM configuration, start the VM, and install a guest OS, refer to the SUSE documentation at SUSE Linux Enterprise Server 15 SP2 Virtualization Guide: Guest Installation.

From a VM configuration perspective, the two most important factors for achieving top performance on CPU and memory bound workloads are:

  1. Partitioning of host resources and placement of the VMs

  2. Enlightenment of the VMs about their virtual topology

For example, if there are two VMs, each one should be run on one socket, to minimize CPU contention and maximize and memory access parallelism. Also, and especially if the VMs are big enough, they should be made aware of their own virtual topology. That way, all the tuning actions described in the first part of this document become applicable to the workloads running inside the VMs too.

15.1 Placement of VMs

When a VM is created, memory is allocated on the host to act as its virtual RAM. Moving this memory, for example on a different NUMA node from the one where it is first allocated, incurs (when possible) in a significant performance impact. Therefore, it is of paramount importance that the initial placement of the VMs is as optimal as possible. Both Xen and KVM can make educated guesses on what a good placement might be. For the purpose of this document, however, we are interested in what is the absolute best possible initial placement of the VMs, taking into account the specific characteristics of AMD EPYC 7003 Series processor systems. And this can only be achieved by manually doing the initial placement. Of course, this comes at the price of reduced flexibility, and is only possible when there is no oversubscription.

Deciding on what pCPUs the vCPUs of a VM run, can be done at creation time, but also easily changed along the VM lifecycle. It is, however, still recommended to start the VMs with good vCPUs placement. This is particularly important on Xen, where vCPU placement is what actually drives and controls memory placement.

Placement of VM memory on host NUMA nodes happens by means of the <numatune> XML element:

<numatune>
  <memory mode='strict' nodeset='0-1'/>
  <memnode cellid='0' mode='strict' nodeset='0'/>
  <memnode cellid='1' mode='strict' nodeset='1'/>
  ...
</numatune>

The 'strict' guarantees that the all memory for the VM will come only from the specified (set of) NUMA node(s) specified in nodeset. A cell is, in fact, a virtual NUMA node, with cellid being its ID and nodeset telling exactly from what host physical NUMA node(s) the memory for this virtual node should be allocated. A NUMA-unaware VMs can still include this element in its configuration file. It will have only one <memnode> element.

Placement of vCPUs happens via the <cputune> element, as in the example below:

<vcpu placement='static'>240</vcpu>
<cputune>
  <vcpupin vcpu='0' cpuset='4'/>
  <vcpupin vcpu='1' cpuset='132'/>
  <vcpupin vcpu='2' cpuset='5'/>
  <vcpupin vcpu='3' cpuset='133'/>
  ...
</cputune>

The value 240 means that the VM will have 240 vCPUs. The static guarantees that each vCPU will stay on the pCPU(s) on which it is pinned to. The various <vcpupin> elements are where the mapping between vCPUs and pCPUs is established (vcpu being the vCPU ID and cpuset being either one or a list of pCPU IDs).

To use all of the available 256 pCPUs for a VM, specify 256. In addition to that, though, the following element should be added in the <device> section:

<devices>
  ...
  <iommu model='intel'>
    <driver intremap='on'/>
  </iommu>
  ...
</devices>

When pinning vCPUs to pCPUs, adjacent vCPU IDs (like vCPU ID 0 and vCPU ID 1) must be assigned to host SMT siblings (like pCPU 65 and pCPU 193, on our test server). In fact, QEMU uses a static SMT sibling CPU ID assignment. This is how we guarantee that virtual SMT siblings will always execute on actual physical SMT siblings.

The following sections give more specific and precise details about placement of vCPUs and memory for VM of varying sizes of VMs, on the reference system for this guide (see Figure 1, “AMD EPYC 7003 Series Processor Topology”.

15.1.1 Placement of only one large VM

An interesting use case when only one VM is used. Reasons for doing something like that include security/isolation, flexibility, high availability, and others. In this cases, typically, the VM is almost as large as the host itself.

Let us, therefore, consider a VM with all of the 256 vCPUs and 200 GB of RAM. Such VM spans multiple host NUMA nodes and therefore it is recommended to create a virtual NUMA topology for it.

We should create a virtual topology with 2 virtual NUMA nodes (that is, as many as there are physical NUMA nodes), and split the VM’s memory equally among them. The 256 vCPUs are assigned to the 2 nodes, 128 on each.

The VM Memory must be split equally among the 2 nodes. At this point, if a suitable virtual topology is also provided to the VM, each of the VM’s vCPUs will access its own memory directly, and use Infinity Fabric inter-socket links only to reach foreign memory, as it happens on the host. Workloads running inside the VM can be tuned exactly like they were running on a bare metal 3rd Generation AMD EPYC processor server.

The following example numactl output comes from a VM configured as explained:

available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
node 0 size: 100670 MB
node 0 free: 100003 MB
node 1 cpus: 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
node 1 size: 100768 MB
node 1 free: 100077 MB
node distances:
node   0   1
  0:  10  32
  1:  32  10

This is analogous to the host topology, also from the point of view of NUMA node distances (as the libvirt version present in SUSE Linux Enterprise Server 15 SP2, allows us to define the virtual nodes distance table). The only differences are the number of CPUs, their APIC IDs and the amount of memory.

See Section 20, “Appendix A” for an almost complete VM configuration file.

An alternative setup would be to leave some pCPUs to the host, for example using only 240 vCPUs for the VM. As said already, full cores must always be used. That means, assign vCPUs 0 and 1 to Core L#4, vCPUs 2 and 3 to Core L#5, vCPUs 4 and 5 to Core L#6, etc. To enact this, we must pin vCPUs 0 and 1 to pCPUs 4 and 132; vCPUs 2 and 3 to pCPUs 5 and 133; vCPUs 4 and 5 to pCPUs 6 and 134, etc. And, of course, the two nodes must have the same number of vCPUs.

In such a setup, cores L#0, L#1, L#2 and L#3 from host NUMA node #0 and cores L#64, L#65, L#66 and L#67 from host NUMA node #1 will be reserved to the host. This, however, means splitting two of the dies, which in turns means there will be some L3 cache interference between the host and the VM. If that damages the performance of the workloads of interest too much, avoid it, by giving to the VM 224 vCPUs.

15.1.2 Placement of two large VMs

If wanting to run two VMs, they can have 128 vCPUs and 100 GB memory each. The recommendation, in this case, is to put each VM on one host NUMA node. This way, VMs no longer span multiple host NUMA nodes, and hence there is no need to define virtual NUMA nodes, as part of their topologies.

Placing each VM on one node means that workloads running inside each of them will never need to use inter-socket interconnect.

In this example scenario, the numactl output of both VMs looks as follows:

available: 1 nodes (0)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
node 0 size: 100572 MB
node 0 free: 100040 MB
node distances:
node   0
  0:  10

15.1.3 Placement of four to sixteen medium-size VMs

Following the same principles, we can split each node in 2, and have 4 VMs with 64 vCPUs and up to 48 GB of RAM each.

Or even have 16 VMs with 16 vCPUs and 12 GB RAM each.

In both cases, VMs will not need to (and must not) be put across the host NUMA nodes boundary, and hence they do not need to be NUMA-aware. They still benefit from having a virtual topology, in terms of cores and threads, which matches the resource they are using. You will find more details about this in following sections.

15.1.4 Placement of many small VMs

If small VMs are the target, VMs with either 8, 4 or 2 vCPUs each can be created and efficiently allocated on AMD EPYC 7003 Series processors servers.

VMs with 8 vCPUs occupies half a dies, and we can have 32 of them. VMs with either 4 or 2 vCPUs will be given 2 and 1 host cores and we can have 64 and 128 of them, respectively. And this is all possible without the need for any of the VMs to span more than one host NUMA node.

Note
Note: Hundreds of VMs with 256 GB of RAM

When the number of VM increases so much, the limiting factor will become the memory. With ~200 GB of RAM available for VMs on our server, we could give only 1 GB of RAM to each of the 128 VMs. Of course, this limitation is only an issue of the system used as reference for this guide. The AMD EPIC 7003 Series Processor architecture itself can accommodate much more RAM.

In cases when it is enough or desirable to have VMs with only 1 vCPUs, it is recommended to still not go beyond 128, and assign no less than one full core to each VM. It should be avoided to have vCPUs from different VMs running on two siblings threads of the same core. Such a solution, although functional, is not ideal for all the cases where good and consistent performance within VMs is a goal. What can be done for each VM and for each core is to use one of the two threads for the actual vCPU, and the other for its IO thread(s), on the host. This solution allows to take advantage of the full 3rd Generation AMD EPYC processing power. However, we advise to check and verify whether this is a good setup and if it brings performance benefits for the specific workloads of interest, before committing to it.

15.2 Emulator IO threads & disaggregation

IO for the VMs is carried out by device emulators running on the host, or by the so-called back-ends of paravirtualized drivers (also running on the host). Both the IO threads of the emulators and the back-ends are seen as (either user or kernel) threads by the host OS. As such, they can be configured to run on specific subsets of the CPUs assigned to the host OS (Dom0’s virtual CPUs, in the case of Xen). And their placement on such CPUs may have an impact on performance of VMs, especially on IO bound workloads.

For example, a 1 vCPU VM can have its vCPU bound to the one SMT thread of a core, while the other thread can be assigned to the host, and the IO thread(s) of the VM pinned to it. The idea behind this is that a common execution pattern for the vCPU is to be idle when the IO threads are busy. Hence this setup potentially maximizes the exploitation of hardware resources.

On Xen, there is also the possibility of setting up driver domains. These are special VMs which act as back-ends of a particular IO device, for one or more VMs. In case they are used, make sure that they run close enough to both the hardware they are providing their abstraction for, and the VMs that they are servicing.

15.3 Oversubscription of host resources

Oversubscription happens when the demand for some resource is higher than the resource is physically available. In virtualization, this can happen for CPU and memory.

Note
Note: Not covering oversubscribed scenarios

This guide does not cover in great details oversubscribed scenarios. However, given the large number of CPUs available on an AMD EPYC 7003 Series processor system and the huge amount of memory the architecture supports, this is not considered a limitation.

CPU Oversubscription

CPU oversubscription happens when the total cumulative number of vCPUs from all VMs becomes higher than 256. Such situation unavoidably introduces latencies, and leads to lower performance than when host resources are just enough. It is, however, impossible to tell a priori by what extent this happens, at least not without a detailed knowledge of the actual workload.

If we knew in advance that the load on each vCPU will always stay below 50%, we could assume that even oversubscribing the host by a factor of 2 (like with ~512 vCPUs, in our case!) will work well. On the other hand, if we knew that the load that each vCPU tries to impose on the system is always 100%, creating even 1 vCPUs more than the host has pCPUs would already lead to a misconfiguration.

When under oversubscription, exclusive 1-to-1 vCPU to pCPU assignment may not the best approach. However, we need to trust the hypervisor scheduler to handle the situation well (which is very likely what will happen). What can be done, though, is to provide the hypervisor scheduler with some suggestions, if we have enough knowledge about the workload. For example, we can try to help the scheduler avoiding that all the VMs running CPU intensive workloads end on the same nodes, or avoiding cross-node VM migrations happening too frequently.

This grouping of VMs on (a set of) nodes can be done with some less strict forms of vCPU pinning (often called affinity), or with other mechanisms. For example, cgroups can be leveraged on a KVM host, or cpupools on a Xen host. Xen also contains a feature, called soft vCPU affinity, which can be used together with traditional vCPU affinity (also called hard vCPU affinity) as a finer grained and more powerful way of controlling resource allocation in oversubscribed scenario.

Memory Oversubscription

Memory oversubscription happens when the total cumulative amount of memory used by all VMs is more than the RAM available on the host. In this case, some of such memory is kept outside of the RAM (swapped out) when the VMs using it are not running. It is put back in RAM (swapped in) when these VMs run again. This also has (potentially severe) performance implications, and is not analyzed in details in here.

On KVM, swapping is handled by Linux virtual memory management and paging code and mechanism, exactly as it happens for host/bare metal processes. On Xen, this is not supported unless special technologies (xenpaging or transcendent memory) are employed.

Then there are page sharing and memory ballooning.

Page sharing relies on the principle that, when VMs use memory pages which are identical, only one copy of them needs to be kept in physical RAM, at least for most of the time. This is supported on KVM, via a mechanism called Kernel Samepage Merging (KSM). On Xen, this again require special tools and actions.

Ballooning relies on the idea that VMs do not always use all the memory they are configured to be able to use. This is supported in both Xen and KVM. However, using it is not ideal for performance, thus it is not analyzed in any further details in this document.

Oversubscription with a Single VM

This is always considered a misconfiguration. In fact, in some more details:

  • A single VM should never have more vCPUs than the host has pCPUs.

  • A single VM should never have more memory than the host has physical RAM.

15.4 Enlightenment of VMs

The term enlightenment refers to letting the guest OS know as many details as possible about the virtual topology of the VM. This is only useful and brings actual performance improvements only if such topology is properly and effectively mapped on host resources, and if such mapping is stable and does not change during the life of the VM.

15.4.1 Virtual CPUs model and topology

To ensure the VM has a vCPU topology, use the following:

<cpu mode='custom' match='exact' check='none'>
  <model fallback='allow'>EPYC</model>
  <topology sockets='2' cores='64' threads='2'/>
  <numa>
    <cell id='0' cpus='0-127' memory='104857600' unit='KiB'>
      <distances>
        <sibling id='0' value='10'/>
        <sibling id='1' value='32'/>
      </distances>
    </cell>
    <cell id='1' cpus='128-255' memory='104857600' unit='KiB'>
      <distances>
        <sibling id='0' value='32'/>
        <sibling id='1' value='10'/>
      </distances>
    </cell>
   </numa>
</cpu>

The <topology> element specifies the CPU characteristics. In this case, we are creating vCPUs which will be seen by the guest OS as being arranged in 2 sockets, each of which has 64 cores, each of which has 2 threads. And this is how we match, for the one big VM, the topology of the host.

Each <cell> element defines one virtual NUMA node, specifying how much memory and which vCPUs are part of it. The <distances> elements allow us to define a custom virtual NUMA distance table. Values should be - as it is on actual hardware - an indication of the latency of accessing the memory of every node from any other. And again we are defining a virtual distance table which exactly matches the host one, for the one big VM.

CPU model is how the VM is told what type of vCPUs should be emulated, for example, what features and special instructions and extensions will be available. To achieve best performance, using mode='host-passthrough' is often the right choice. However, at least on SUSE Linux Enterprise Server 15 SP2, this produces wrong results. In fact, such setting prevents the topology - despite it being described correctly in the <topology> element - from being interpreted correctly, inside the VM. This is visible in the output of the command lscpu:

...
CPU(s):              256
On-line CPU(s) list: 0-255
Thread(s) per core:  1
Core(s) per socket:  128
Socket(s):           2
NUMA node(s):        2
...
L1d cache:           64K
L1i cache:           64K
L2 cache:            512K
L3 cache:            16384K
NUMA node0 CPU(s):   0-127
NUMA node1 CPU(s):   128-255
...

Benchmarks have shown that misunderstanding the CPUs as being all individual cores, without there being SMT, has a negative impact on performance. Note that the problem not only involves CPU topology, it is also the cache hierarchy structure which is not constructed consistently with the host. In fact, while the sizes of L1d and L1i caches is wrong (it is 64k instead than 32k, see Figure 1, “AMD EPYC 7003 Series Processor Topology”). It is also the information about which CPUs share what caches that is completely misconfigured, as shown here:

cat /sys/devices/system/cpu/cpu0/cache/index0/shared_cpu_list
0
cat /sys/devices/system/cpu/cpu0/cache/index1/shared_cpu_list
0
cat /sys/devices/system/cpu/cpu0/cache/index2/shared_cpu_list
0
cat /sys/devices/system/cpu/cpu0/cache/index3/shared_cpu_list
0-128

The VM thinks that L1d, L1i and L2 are private of each vCPU. That is not the case, as they are shared by the two threads of each core. Even worse, it thinks that L3 is shared across all the vCPUs of a socket, which is not true, as it is shared only among 8 cores (that is 16 vCPUs).

For this reason, the recommendation is to use EPYC as the CPU model (as a 3rd Generation AMD EPYC processor model is not yet available). This enables a virtual topology inside the VM what is more consistent (although not perfectly identical) with the one of the host, and benchmark shows a beneficial effect on performance coming from that. The output of lscpu with EPYC as a model is shown below:

...
CPU(s):              256
On-line CPU(s) list: 0-255
Thread(s) per core:  2
Core(s) per socket:  64
Socket(s):           2
NUMA node(s):        2
...
L1d cache:           32K
L1i cache:           64K
L2 cache:            512K
L3 cache:            8192K
NUMA node0 CPU(s):   0-127
NUMA node1 CPU(s):   128-255
...

Now it is the cache sizes that are slightly off, with regard to the host value. And this is because they are the values taken from previous generation AMD EPYC processors architecture characteristics. But the cache topology and cache level sharing among vCPUs are now closer to the host ones (not perfectly, as index3 should report 0-15). This is far more important for achieving performance inside the VM that matches the one of the host. This is shown here:

cat /sys/devices/system/cpu/cpu0/cache/index0/shared_cpu_list
0-1
cat /sys/devices/system/cpu/cpu0/cache/index1/shared_cpu_list
0-1
cat /sys/devices/system/cpu/cpu0/cache/index2/shared_cpu_list
0-1
cat /sys/devices/system/cpu/cpu0/cache/index3/shared_cpu_list
0-7

Related to this, on SUSE Linux Enterprise Server 15 SP2 it would be possible to use the haltpoll CPU-Idle governor, inside the VM. This is an optimization meant at reducing the number of context switches from the VM and the host (also known as VMExits), when doing static resource partitioning of the host itself. It is would therefore be recommended here, but the problem is that it can be enabled only if the CPU model used is host-passthrough.

We therefore recommend to not use it in a SUSE Linux Enterprise Server 15 SP2 VM running on an AMD EPYC 7003 Series processors, until the inaccurate vCPU topology problem is resolved (and an investigation is ongoing for figuring out how to best address the problem).

15.4.2 Memory backing

Memory wise, the VMs need to be told to use the Huge Pages that were reserved for them. To effectively use 1 GB huge pages, the amount of memory each VM is given must be a multiple of 1 GB. Also, KSM should be disabled and we also must ensure that VMs' memory is never going to be swapped out. This is all achieved as follows:

<memory unit='KiB'>209715200</memory>
<memoryBacking>
  <hugepages>
    <page size='209715200' unit='KiB'/>
  </hugepages>
  <nosharepages/>
  <locked/>
</memoryBacking>
Note
Note: Huge Pages, shared pages and locked pages on Xen

On Xen, for HVM guests, huge pages are used by default and there is neither any page sharing nor swapping (at least not in SUSE Linux Enterprise Server 15 SP1). Therefore, the entire <memoryBacking> element is technically not necessary.

15.4.3 Ballooning

To get the full benefit of Huge Pages, memory ballooning must also be disabled. In fact, if the ballooning driver is not Huge Pages-aware, using ballooning would end up splitting the pages in fragments, neutering their (positive) performance impact. Disabling memory ballooning is done as follows:

<memory unit='KiB'>209715200</memory>
<currentMemory unit='KiB'>209715200</currentMemory>

That is, by specifying in the <currentMemory> the same value already used in the <memory> element (and never changing the memory size of the VM at runtime).

15.4.4 (Transparent) Huge Pages

If huge pages are used for allocating the VMs’ memory on the host, they can also be used inside the VMs, either explicitly, or via THP. Whether that helps performance is workload dependent. The analysis and the considerations made in the first part of the document about using (T)HP on bare metal, can also be applied here.

15.4.5 Automatic NUMA balancing

Similarly to THP, if the VM is NUMA-aware, NUMAB can be used inside of it to boost the performance of most NUMA-unaware workloads running inside the VM itself.

15.4.6 Services and daemons

irqbalance can be a source of latency inside of the VM, because of the way it uses the /proc/interrupts interface. For workloads that are particularly sensitive to latency, consider disabling it within the VMs (of course by taking the appropriate alternative measures, like binding IRQs, if necessary).

15.5 Secure Encrypted Virtualization

The following documents

provide details on how to configure a VM to use AMD SEV (on an AMD SEV capable and properly configured host). Therefore, this section will not go into details. It only lists the entries that are required to add to or change the configuration file.

First of all, a <launchSecurity> stanza must be added:

<launchSecurity type='sev'>
  <cbitpos>51</cbitpos>
  <reducedPhysBits>1</reducedPhysBits>
  <policy>0x0003</policy>
</launchSecurity>

Two other things are that the QEMU machine type used for the VM should be Q35 and that a UEFI must be used:

<os>
  <type arch='x86_64' machine='pc-q35-3.1'>hvm</type>
  <loader readonly='yes' type='pflash'>/usr/share/qemu/ovmf-x86_64-ms-4m-code.bin</loader>
  <nvram>/var/lib/libvirt/qemu/nvram/sle15sp1-vm1-sev_VARS.fd</nvram>
</os>

Note that on SUSE Linux Enterprise Server 15 SP2, an SEV-enabled VM can use 1 GB Huge Pages memory, preallocated at boot time, the same as when SEV is not enabled as they are allocated during boot.

Finally, any device that uses 'virtio' needs to gain an XML element looking like: <driver iommu='on'/>

At this point, we can create and start a VM with its memory encrypted, through AMD SEV technology:

dmesg |grep SEV
[    0.004000] AMD Secure Encrypted Virtualization (SEV) active

SEV and The IOMMU Device

In our experiments, for having a functional VM with SEV enabled, we had to remove the iommu element, in the <device> section of the VM configuration. Since that element was necessary for having more than 255 vCPUs in the VM, all the experiments conducted with SEV enabled have been done in VMs with at most 248 vCPUs.

16 Test VM workload: STREAM

The effectiveness of the proposed tuning is demonstrated using the STREAM benchmark.

16.1 Test scenario: One large VM

Figure 4, “STREAM Bandwidth - Bare metal compared with one VM” shows the bandwidth achieved by STREAM when the benchmark is built in single thread mode (single) or parallelized with OpenMP (omp-llcs-spread). For both cases, we compare the results achieved on the host (BM) and inside one VM (VM).

STREAM Bandwidth - Bare metal compared with one VM
Figure 4: STREAM Bandwidth - Bare metal compared with one VM

As the VM has approximately the same dimension and topology of the host, the same STREAM configuration parameters used in Section 12.2, “Test workload: STREAM”, are used also here.

We can clearly see how proper tuning allows a VM running on an AMD EPYC 7003 Series processor server to achieve a memory bandwidth performance within 1% of the one of the host.

Figure 5, “STREAM Bandwidth - Single thread in one VM with different CPU models” and Figure 4, “STREAM Bandwidth - Bare metal compared with one VM” show the effect of using different CPU models for the virtual machine. While with single threaded STREAM it seems that CPU model is not a really important parameter, the OpenMP run clearly show how the using the host-passthrough dramatically decreases the performance.

This seems counter-intuitive, as using host-passthrough and enabling the haltpoll governor inside the guest is meant to provide the absolute best performance when doing static host resource partitioning, as we are doing here. However, we know that in our case this is because of the inaccurate cache hierarchy exposed to the VM that this causes.

STREAM Bandwidth - Single thread in one VM with different CPU models
Figure 5: STREAM Bandwidth - Single thread in one VM with different CPU models
STREAM Bandwidth - OpenMP in one VM with different CPU models
Figure 6: STREAM Bandwidth - OpenMP in one VM with different CPU models

This confirms the need to always run benchmarks for assessing the performance of the relevant workloads, on a given platform. In fact, because of specific characteristics of some components of the virtualization stack available in SUSE Linux Enterprise Server 15 SP2, the EPYC CPU model (host-model) is preferable to what it would have appeared to be the most obvious choice, which is the host-passthrough. CPU model with haltpoll governor enabled inside the guest.

16.2 Test scenario: Multiple VMs

Section 16.3, “Test scenario: Secure Encrypted Virtualization” shows what happens when two or four large VMs are used, both with the recommended tuning applied (performance of 1 VM is also displayed, for reference). The graph reports the average memory bandwidth achieved in each experiments, considering all the VMs involved. For example, the light blue bar (4 VMS) is relative to an experiment where four VMs were used and represents the average of the four different memory bandwidth values, one for each of the VMs.

STREAM Bandwidth - OpenMP, per-VM average when using multiple VMs
Figure 7: STREAM Bandwidth - OpenMP, per-VM average when using multiple VMs

We can see that performance scales down in a close to linear fashion, with the increase of the number of VMs involved (although, not perfectly linear for all the type of STREAM operations).

16.3 Test scenario: Secure Encrypted Virtualization

The goal for this test was evaluating the impact on the performance of SEV, on a memory intensive workload like STREAM.

We can evaluate that looking at the results of running single threaded and parallelized STREAM in one VM (in Figure 8, “STREAM Bandwidth - Single and OpenMP, with and without SEV in 1 VM”), in two VMs (in Figure 9, “STREAM Bandwidth - Single and OpenMP, with and without SEV in 2 VMs”) and in four VMs *in Figure 10, “NAS Completion Time - Baremetal and in one VM, different configurations”) running concurrently, both when SEV is disabled (no-SEV) and when it is enabled (SEV). They again report the average of the bandwidth values of each VM that was running in the particular experiment.

STREAM Bandwidth - Single and OpenMP, with and without SEV in 1 VM
Figure 8: STREAM Bandwidth - Single and OpenMP, with and without SEV in 1 VM
STREAM Bandwidth - Single and OpenMP, with and without SEV in 2 VMs
Figure 9: STREAM Bandwidth - Single and OpenMP, with and without SEV in 2 VMs
NAS Completion Time - Baremetal and in one VM, different configurations
Figure 10: NAS Completion Time - Baremetal and in one VM, different configurations

It is quite evident that enabling SEV has only a very limited impact on the performance for this workload, which is within 2% and 4% for most of the experiments. Even in the case where the overhead is the highest (OpenMP Stream run inside 4 VMs), the performance decrease is limited to around 6%.

17 Test VM workload: NPB

To match the experimental evaluation done on bare-metal, NPB base workloads have also been considered.

17.1 Test scenario: One large VM

Figure 11, “NAS Completion Time - Baremetal and in one VM with different configurations” shows the results of running the various NPB workloads on the host in the default (BM default), tuned (BM tuned) and selective (BM selective) configurations, compared with the exact same done inside one VM (VM default, VM tuned, VM selective).

NAS Completion Time - Baremetal and in one VM with different configurations
Figure 11: NAS Completion Time - Baremetal and in one VM with different configurations

This makes pretty evident that, with proper tuning, also CPU bound workload running inside a VM can achieve host-level performance.

Figure 12, “NAS Completion Time - Selective configuration in one VM, varying the CPU model” shows the impact of selecting a specific CPU model for the VM (only the results for the selective configuration are reported). We see that this aspect seems to play a smaller role for this benchmark, as compared to how it changed the performance for STREAM. In any case, even for CPU bound workloads like NAS, the EPYC model is the one to be preferred.

NAS Completion Time - Selective configuration in one VM, varying the CPU model
Figure 12: NAS Completion Time - Selective configuration in one VM, varying the CPU model

17.2 Test scenario: Multiple VMs with Secure Encrypted Virtualization

Similarly to what has been done with STREAM, we evaluate here the overhead introduced by SEV for CPU on the NPB workloads and we do that while running it inside two (in Figure 13, “NAS Completion Time - Tuned configuration in 2 VM, with and without SEV”) and in four (Figure 14, “NAS Completion Time - Tuned configuration in 4 VM, with and without SEV”) VMs, with both SEV off and on.

NAS Completion Time - Tuned configuration in 2 VM, with and without SEV
Figure 13: NAS Completion Time - Tuned configuration in 2 VM, with and without SEV
NAS Completion Time - Tuned configuration in 4 VM, with and without SEV
Figure 14: NAS Completion Time - Tuned configuration in 4 VM, with and without SEV

Results are again pretty good, as the performance penalty due to SEV, on a CPU intensive benchmark like NPB, is nearly non-existent (it is less than 0.5%).

18 Conclusion

The introduction of AMD EPYC 7003 Series processors pushes the boundaries of what is possible for memory and IO-bound workloads with much higher bandwidth and available number of channels. A properly configured and tuned workload can exceed the performance of many contemporary off-the-shelf solutions even when fully customized. The symmetric and balanced nature of the machine makes the task of tuning a workload considerably easier given that each partition can have symmetric performance. And this is a property that can turn out particularly handy in virtualization, as each virtual machine can be assigned to each one of the partitions.

With SUSE Linux Enterprise, all the tools to monitor and tune a workload are readily available. Your customers can extract the maximum performance and reliability running their applications on the 3rd Generation AMD EPYC processor platform.

19 Resources

For more information, refer to:

20 Appendix A

Example of a VM configuration file:

<domain type="kvm">
  <name>vm1</name>
  <uuid>e9afa48f-b359-4389-9ae6-5589240867c0</uuid>
  <metadata>
    <libosinfo:libosinfo xmlns:libosinfo="http://libosinfo.org/xmlns/libvirt/domain/1.0">
      <libosinfo:os id="http://suse.com/sle/15.2"/>
    </libosinfo:libosinfo>
  </metadata>
  <memory unit="KiB">209715200</memory>
  <currentMemory unit="KiB">209715200</currentMemory>
  <memoryBacking>
    <hugepages>
      <page size="1048576" unit="KiB"/>
    </hugepages>
    <nosharepages/>
    <locked/>
  </memoryBacking>
  <vcpu placement="static">256</vcpu>
  <cputune>
    <vcpupin vcpu="0" cpuset="0"/>
    <vcpupin vcpu="1" cpuset="128"/>
    <vcpupin vcpu="2" cpuset="1"/>
    <vcpupin vcpu="3" cpuset="129"/>
    <vcpupin vcpu="4" cpuset="2"/>
    <vcpupin vcpu="5" cpuset="130"/>
    <vcpupin vcpu="6" cpuset="3"/>
    <vcpupin vcpu="7" cpuset="131"/>
    <vcpupin vcpu="8" cpuset="4"/>
    <vcpupin vcpu="9" cpuset="132"/>
    <vcpupin vcpu="10" cpuset="5"/>
    <vcpupin vcpu="11" cpuset="133"/>
    <vcpupin vcpu="12" cpuset="6"/>
    <vcpupin vcpu="13" cpuset="134"/>
    <vcpupin vcpu="14" cpuset="7"/>
    <vcpupin vcpu="15" cpuset="135"/>
    <vcpupin vcpu="16" cpuset="8"/>
    <vcpupin vcpu="17" cpuset="136"/>
    <vcpupin vcpu="18" cpuset="9"/>
    <vcpupin vcpu="19" cpuset="137"/>
    <vcpupin vcpu="20" cpuset="10"/>
    <vcpupin vcpu="21" cpuset="138"/>
    <vcpupin vcpu="22" cpuset="11"/>
    <vcpupin vcpu="23" cpuset="139"/>
    <vcpupin vcpu="24" cpuset="12"/>
    <vcpupin vcpu="25" cpuset="140"/>
    <vcpupin vcpu="26" cpuset="13"/>
    <vcpupin vcpu="27" cpuset="141"/>
    <vcpupin vcpu="28" cpuset="14"/>
    <vcpupin vcpu="29" cpuset="142"/>
    <vcpupin vcpu="30" cpuset="15"/>
    <vcpupin vcpu="31" cpuset="143"/>
    <vcpupin vcpu="32" cpuset="16"/>
    <vcpupin vcpu="33" cpuset="144"/>
    <vcpupin vcpu="34" cpuset="17"/>
    <vcpupin vcpu="35" cpuset="145"/>
    <vcpupin vcpu="36" cpuset="18"/>
    <vcpupin vcpu="37" cpuset="146"/>
    <vcpupin vcpu="38" cpuset="19"/>
    <vcpupin vcpu="39" cpuset="147"/>
    <vcpupin vcpu="40" cpuset="20"/>
    <vcpupin vcpu="41" cpuset="148"/>
    <vcpupin vcpu="42" cpuset="21"/>
    <vcpupin vcpu="43" cpuset="149"/>
    <vcpupin vcpu="44" cpuset="22"/>
    <vcpupin vcpu="45" cpuset="150"/>
    <vcpupin vcpu="46" cpuset="23"/>
    <vcpupin vcpu="47" cpuset="151"/>
    <vcpupin vcpu="48" cpuset="24"/>
    <vcpupin vcpu="49" cpuset="152"/>
    <vcpupin vcpu="50" cpuset="25"/>
    <vcpupin vcpu="51" cpuset="153"/>
    <vcpupin vcpu="52" cpuset="26"/>
    <vcpupin vcpu="53" cpuset="154"/>
    <vcpupin vcpu="54" cpuset="27"/>
    <vcpupin vcpu="55" cpuset="155"/>
    <vcpupin vcpu="56" cpuset="28"/>
    <vcpupin vcpu="57" cpuset="156"/>
    <vcpupin vcpu="58" cpuset="29"/>
    <vcpupin vcpu="59" cpuset="157"/>
    <vcpupin vcpu="60" cpuset="30"/>
    <vcpupin vcpu="61" cpuset="158"/>
    <vcpupin vcpu="62" cpuset="31"/>
    <vcpupin vcpu="63" cpuset="159"/>
    <vcpupin vcpu="64" cpuset="32"/>
    <vcpupin vcpu="65" cpuset="160"/>
    <vcpupin vcpu="66" cpuset="33"/>
    <vcpupin vcpu="67" cpuset="161"/>
    <vcpupin vcpu="68" cpuset="34"/>
    <vcpupin vcpu="69" cpuset="162"/>
    <vcpupin vcpu="70" cpuset="35"/>
    <vcpupin vcpu="71" cpuset="163"/>
    <vcpupin vcpu="72" cpuset="36"/>
    <vcpupin vcpu="73" cpuset="164"/>
    <vcpupin vcpu="74" cpuset="37"/>
    <vcpupin vcpu="75" cpuset="165"/>
    <vcpupin vcpu="76" cpuset="38"/>
    <vcpupin vcpu="77" cpuset="166"/>
    <vcpupin vcpu="78" cpuset="39"/>
    <vcpupin vcpu="79" cpuset="167"/>
    <vcpupin vcpu="80" cpuset="40"/>
    <vcpupin vcpu="81" cpuset="168"/>
    <vcpupin vcpu="82" cpuset="41"/>
    <vcpupin vcpu="83" cpuset="169"/>
    <vcpupin vcpu="84" cpuset="42"/>
    <vcpupin vcpu="85" cpuset="170"/>
    <vcpupin vcpu="86" cpuset="43"/>
    <vcpupin vcpu="87" cpuset="171"/>
    <vcpupin vcpu="88" cpuset="44"/>
    <vcpupin vcpu="89" cpuset="172"/>
    <vcpupin vcpu="90" cpuset="45"/>
    <vcpupin vcpu="91" cpuset="173"/>
    <vcpupin vcpu="92" cpuset="46"/>
    <vcpupin vcpu="93" cpuset="174"/>
    <vcpupin vcpu="94" cpuset="47"/>
    <vcpupin vcpu="95" cpuset="175"/>
    <vcpupin vcpu="96" cpuset="48"/>
    <vcpupin vcpu="97" cpuset="176"/>
    <vcpupin vcpu="98" cpuset="49"/>
    <vcpupin vcpu="99" cpuset="177"/>
    <vcpupin vcpu="100" cpuset="50"/>
    <vcpupin vcpu="101" cpuset="178"/>
    <vcpupin vcpu="102" cpuset="51"/>
    <vcpupin vcpu="103" cpuset="179"/>
    <vcpupin vcpu="104" cpuset="52"/>
    <vcpupin vcpu="105" cpuset="180"/>
    <vcpupin vcpu="106" cpuset="53"/>
    <vcpupin vcpu="107" cpuset="181"/>
    <vcpupin vcpu="108" cpuset="54"/>
    <vcpupin vcpu="109" cpuset="182"/>
    <vcpupin vcpu="110" cpuset="55"/>
    <vcpupin vcpu="111" cpuset="183"/>
    <vcpupin vcpu="112" cpuset="56"/>
    <vcpupin vcpu="113" cpuset="184"/>
    <vcpupin vcpu="114" cpuset="57"/>
    <vcpupin vcpu="115" cpuset="185"/>
    <vcpupin vcpu="116" cpuset="58"/>
    <vcpupin vcpu="117" cpuset="186"/>
    <vcpupin vcpu="118" cpuset="59"/>
    <vcpupin vcpu="119" cpuset="187"/>
    <vcpupin vcpu="120" cpuset="60"/>
    <vcpupin vcpu="121" cpuset="188"/>
    <vcpupin vcpu="122" cpuset="61"/>
    <vcpupin vcpu="123" cpuset="189"/>
    <vcpupin vcpu="124" cpuset="62"/>
    <vcpupin vcpu="125" cpuset="190"/>
    <vcpupin vcpu="126" cpuset="63"/>
    <vcpupin vcpu="127" cpuset="191"/>
    <vcpupin vcpu="128" cpuset="64"/>
    <vcpupin vcpu="129" cpuset="192"/>
    <vcpupin vcpu="130" cpuset="65"/>
    <vcpupin vcpu="131" cpuset="193"/>
    <vcpupin vcpu="132" cpuset="66"/>
    <vcpupin vcpu="133" cpuset="194"/>
    <vcpupin vcpu="134" cpuset="67"/>
    <vcpupin vcpu="135" cpuset="195"/>
    <vcpupin vcpu="136" cpuset="68"/>
    <vcpupin vcpu="137" cpuset="196"/>
    <vcpupin vcpu="138" cpuset="69"/>
    <vcpupin vcpu="139" cpuset="197"/>
    <vcpupin vcpu="140" cpuset="70"/>
    <vcpupin vcpu="141" cpuset="198"/>
    <vcpupin vcpu="142" cpuset="71"/>
    <vcpupin vcpu="143" cpuset="199"/>
    <vcpupin vcpu="144" cpuset="72"/>
    <vcpupin vcpu="145" cpuset="200"/>
    <vcpupin vcpu="146" cpuset="73"/>
    <vcpupin vcpu="147" cpuset="201"/>
    <vcpupin vcpu="148" cpuset="74"/>
    <vcpupin vcpu="149" cpuset="202"/>
    <vcpupin vcpu="150" cpuset="75"/>
    <vcpupin vcpu="151" cpuset="203"/>
    <vcpupin vcpu="152" cpuset="76"/>
    <vcpupin vcpu="153" cpuset="204"/>
    <vcpupin vcpu="154" cpuset="77"/>
    <vcpupin vcpu="155" cpuset="205"/>
    <vcpupin vcpu="156" cpuset="78"/>
    <vcpupin vcpu="157" cpuset="206"/>
    <vcpupin vcpu="158" cpuset="79"/>
    <vcpupin vcpu="159" cpuset="207"/>
    <vcpupin vcpu="160" cpuset="80"/>
    <vcpupin vcpu="161" cpuset="208"/>
    <vcpupin vcpu="162" cpuset="81"/>
    <vcpupin vcpu="163" cpuset="209"/>
    <vcpupin vcpu="164" cpuset="82"/>
    <vcpupin vcpu="165" cpuset="210"/>
    <vcpupin vcpu="166" cpuset="83"/>
    <vcpupin vcpu="167" cpuset="211"/>
    <vcpupin vcpu="168" cpuset="84"/>
    <vcpupin vcpu="169" cpuset="212"/>
    <vcpupin vcpu="170" cpuset="85"/>
    <vcpupin vcpu="171" cpuset="213"/>
    <vcpupin vcpu="172" cpuset="86"/>
    <vcpupin vcpu="173" cpuset="214"/>
    <vcpupin vcpu="174" cpuset="87"/>
    <vcpupin vcpu="175" cpuset="215"/>
    <vcpupin vcpu="176" cpuset="88"/>
    <vcpupin vcpu="177" cpuset="216"/>
    <vcpupin vcpu="178" cpuset="89"/>
    <vcpupin vcpu="179" cpuset="217"/>
    <vcpupin vcpu="180" cpuset="90"/>
    <vcpupin vcpu="181" cpuset="218"/>
    <vcpupin vcpu="182" cpuset="91"/>
    <vcpupin vcpu="183" cpuset="219"/>
    <vcpupin vcpu="184" cpuset="92"/>
    <vcpupin vcpu="185" cpuset="220"/>
    <vcpupin vcpu="186" cpuset="93"/>
    <vcpupin vcpu="187" cpuset="221"/>
    <vcpupin vcpu="188" cpuset="94"/>
    <vcpupin vcpu="189" cpuset="222"/>
    <vcpupin vcpu="190" cpuset="95"/>
    <vcpupin vcpu="191" cpuset="223"/>
    <vcpupin vcpu="192" cpuset="96"/>
    <vcpupin vcpu="193" cpuset="224"/>
    <vcpupin vcpu="194" cpuset="97"/>
    <vcpupin vcpu="195" cpuset="225"/>
    <vcpupin vcpu="196" cpuset="98"/>
    <vcpupin vcpu="197" cpuset="226"/>
    <vcpupin vcpu="198" cpuset="99"/>
    <vcpupin vcpu="199" cpuset="227"/>
    <vcpupin vcpu="200" cpuset="100"/>
    <vcpupin vcpu="201" cpuset="228"/>
    <vcpupin vcpu="202" cpuset="101"/>
    <vcpupin vcpu="203" cpuset="229"/>
    <vcpupin vcpu="204" cpuset="102"/>
    <vcpupin vcpu="205" cpuset="230"/>
    <vcpupin vcpu="206" cpuset="103"/>
    <vcpupin vcpu="207" cpuset="231"/>
    <vcpupin vcpu="208" cpuset="104"/>
    <vcpupin vcpu="209" cpuset="232"/>
    <vcpupin vcpu="210" cpuset="105"/>
    <vcpupin vcpu="211" cpuset="233"/>
    <vcpupin vcpu="212" cpuset="106"/>
    <vcpupin vcpu="213" cpuset="234"/>
    <vcpupin vcpu="214" cpuset="107"/>
    <vcpupin vcpu="215" cpuset="235"/>
    <vcpupin vcpu="216" cpuset="108"/>
    <vcpupin vcpu="217" cpuset="236"/>
    <vcpupin vcpu="218" cpuset="109"/>
    <vcpupin vcpu="219" cpuset="237"/>
    <vcpupin vcpu="220" cpuset="110"/>
    <vcpupin vcpu="221" cpuset="238"/>
    <vcpupin vcpu="222" cpuset="111"/>
    <vcpupin vcpu="223" cpuset="239"/>
    <vcpupin vcpu="224" cpuset="112"/>
    <vcpupin vcpu="225" cpuset="240"/>
    <vcpupin vcpu="226" cpuset="113"/>
    <vcpupin vcpu="227" cpuset="241"/>
    <vcpupin vcpu="228" cpuset="114"/>
    <vcpupin vcpu="229" cpuset="242"/>
    <vcpupin vcpu="230" cpuset="115"/>
    <vcpupin vcpu="231" cpuset="243"/>
    <vcpupin vcpu="232" cpuset="116"/>
    <vcpupin vcpu="233" cpuset="244"/>
    <vcpupin vcpu="234" cpuset="117"/>
    <vcpupin vcpu="235" cpuset="245"/>
    <vcpupin vcpu="236" cpuset="118"/>
    <vcpupin vcpu="237" cpuset="246"/>
    <vcpupin vcpu="238" cpuset="119"/>
    <vcpupin vcpu="239" cpuset="247"/>
    <vcpupin vcpu="240" cpuset="120"/>
    <vcpupin vcpu="241" cpuset="248"/>
    <vcpupin vcpu="242" cpuset="121"/>
    <vcpupin vcpu="243" cpuset="249"/>
    <vcpupin vcpu="244" cpuset="122"/>
    <vcpupin vcpu="245" cpuset="250"/>
    <vcpupin vcpu="246" cpuset="123"/>
    <vcpupin vcpu="247" cpuset="251"/>
    <vcpupin vcpu="248" cpuset="124"/>
    <vcpupin vcpu="249" cpuset="252"/>
    <vcpupin vcpu="250" cpuset="125"/>
    <vcpupin vcpu="251" cpuset="253"/>
    <vcpupin vcpu="252" cpuset="126"/>
    <vcpupin vcpu="253" cpuset="254"/>
    <vcpupin vcpu="254" cpuset="127"/>
    <vcpupin vcpu="255" cpuset="255"/>
  </cputune>
  <numatune>
    <memory mode="strict" nodeset="0-1"/>
    <memnode cellid="0" mode="strict" nodeset="0"/>
    <memnode cellid="1" mode="strict" nodeset="1"/>
  </numatune>
  <os>
    <type arch="x86_64" machine="pc-q35-4.2">hvm</type>
    <loader readonly="yes" type="pflash">/usr/share/qemu/ovmf-x86_64-ms-4m-code.bin</loader>
    <nvram>/var/lib/libvirt/qemu/nvram/vm1_VARS.fd</nvram>
    <boot dev="hd"/>
  </os>
  <features>
    <acpi/>
    <apic/>
    <pae/>
    <vmport state="off"/>
    <ioapic driver="qemu"/>
  </features>
  <cpu mode="custom" match="exact" check="none">
    <model fallback="allow">EPYC</model>
    <topology sockets="2" cores="64" threads="2"/>
    <numa>
      <cell id="0" cpus="0-127" memory="104857600" unit="KiB">
        <distances>
          <sibling id="0" value="10"/>
          <sibling id="1" value="32"/>
        </distances>
      </cell>
      <cell id="1" cpus="128-255" memory="104857600" unit="KiB">
        <distances>
          <sibling id="0" value="32"/>
          <sibling id="1" value="10"/>
        </distances>
      </cell>
    </numa>
  </cpu>
  <clock offset="utc">
    <timer name="rtc" tickpolicy="catchup"/>
    <timer name="pit" tickpolicy="delay"/>
    <timer name="hpet" present="no"/>
  </clock>
  <on_poweroff>destroy</on_poweroff>
  <on_reboot>restart</on_reboot>
  <on_crash>destroy</on_crash>
  <pm>
    <suspend-to-mem enabled="no"/>
    <suspend-to-disk enabled="no"/>
  </pm>
  <devices>
    <emulator>/usr/bin/qemu-system-x86_64</emulator>
    <disk type="file" device="disk">
      <driver name="qemu" type="qcow2"/>
      <source file="/var/lib/libvirt/images/vm1.qcow2"/>
      <target dev="vda" bus="virtio"/>
      <address type="pci" domain="0x0000" bus="0x05" slot="0x00" function="0x0"/>
    </disk>
    ...
    <interface type="network">
      <mac address="52:54:00:16:1e:01"/>
      <source network="default"/>
      <model type="virtio"/>
      <rom enabled="no"/>
      <address type="pci" domain="0x0000" bus="0x01" slot="0x00" function="0x0"/>
    </interface>
    <serial type="pty">
      <target type="isa-serial" port="0">
        <model name="isa-serial"/>
      </target>
    </serial>
    <console type="pty">
      <target type="serial" port="0"/>
    </console>
    ...
    <graphics type="spice" autoport="yes">
      <listen type="address"/>
    </graphics>
    <video>
      <model type="qxl" ram="65536" vram="65536" vgamem="16384" heads="1" primary="yes"/>
      <address type="pci" domain="0x0000" bus="0x00" slot="0x01" function="0x0"/>
    </video>
    <memballoon model="virtio">
      <address type="pci" domain="0x0000" bus="0x06" slot="0x00" function="0x0"/>
    </memballoon>
    <rng model="virtio">
      <backend model="random">/dev/urandom</backend>
      <address type="pci" domain="0x0000" bus="0x08" slot="0x00" function="0x0"/>
    </rng>
    <iommu model="intel">
      <driver intremap="on" eim="on"/>
    </iommu>
    <vsock model="virtio">
      <cid auto="yes"/>
      <address type="pci" domain="0x0000" bus="0x07" slot="0x00" function="0x0"/>
    </vsock>
  </devices>
</domain>

21 Legal notice

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.

SUSE, the SUSE logo and YaST are registered trademarks of SUSE LLC in the United States and other countries. For SUSE trademarks, see http://www.suse.com/company/legal/. Linux is a registered trademark of Linus Torvalds. All other names or trademarks mentioned in this document may be trademarks or registered trademarks of their respective owners.

Documents published as part of the SUSE Best Practices series have been contributed voluntarily by SUSE employees and third parties. They are meant to serve as examples of how particular actions can be performed. They have been compiled with utmost attention to detail. However, this does not guarantee complete accuracy. SUSE cannot verify that actions described in these documents do what is claimed or whether actions described have unintended consequences. SUSE LLC, its affiliates, the authors, and the translators may not be held liable for possible errors or the consequences thereof.

Below we draw your attention to the license under which the articles are published.

22 GNU Free Documentation License

Copyright (C) 2000, 2001, 2002 Free Software Foundation, Inc. 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed.

The purpose of this License is to make a manual, textbook, or other functional and useful document "free" in the sense of freedom: to assure everyone the effective freedom to copy and redistribute it, with or without modifying it, either commercially or non-commercially. Secondarily, this License preserves for the author and publisher a way to get credit for their work, while not being considered responsible for modifications made by others.

This License is a kind of "copyleft", which means that derivative works of the document must themselves be free in the same sense. It complements the GNU General Public License, which is a copyleft license designed for free software.

We have designed this License to use it for manuals for free software, because free software needs free documentation: a free program should come with manuals providing the same freedoms that the software does. But this License is not limited to software manuals; it can be used for any textual work, regardless of subject matter or whether it is published as a printed book. We recommend this License principally for works whose purpose is instruction or reference.

This License applies to any manual or other work, in any medium, that contains a notice placed by the copyright holder saying it can be distributed under the terms of this License. Such a notice grants a world-wide, royalty-free license, unlimited in duration, to use that work under the conditions stated herein. The "Document", below, refers to any such manual or work. Any member of the public is a licensee, and is addressed as "you". You accept the license if you copy, modify or distribute the work in a way requiring permission under copyright law.

A "Modified Version" of the Document means any work containing the Document or a portion of it, either copied verbatim, or with modifications and/or translated into another language.

A "Secondary Section" is a named appendix or a front-matter section of the Document that deals exclusively with the relationship of the publishers or authors of the Document to the Document's overall subject (or to related matters) and contains nothing that could fall directly within that overall subject. (Thus, if the Document is in part a textbook of mathematics, a Secondary Section may not explain any mathematics.) The relationship could be a matter of historical connection with the subject or with related matters, or of legal, commercial, philosophical, ethical or political position regarding them.

The "Invariant Sections" are certain Secondary Sections whose titles are designated, as being those of Invariant Sections, in the notice that says that the Document is released under this License. If a section does not fit the above definition of Secondary then it is not allowed to be designated as Invariant. The Document may contain zero Invariant Sections. If the Document does not identify any Invariant Sections then there are none.

The "Cover Texts" are certain short passages of text that are listed, as Front-Cover Texts or Back-Cover Texts, in the notice that says that the Document is released under this License. A Front-Cover Text may be at most 5 words, and a Back-Cover Text may be at most 25 words.

A "Transparent" copy of the Document means a machine-readable copy, represented in a format whose specification is available to the general public, that is suitable for revising the document straightforwardly with generic text editors or (for images composed of pixels) generic paint programs or (for drawings) some widely available drawing editor, and that is suitable for input to text formatters or for automatic translation to a variety of formats suitable for input to text formatters. A copy made in an otherwise Transparent file format whose markup, or absence of markup, has been arranged to thwart or discourage subsequent modification by readers is not Transparent. An image format is not Transparent if used for any substantial amount of text. A copy that is not "Transparent" is called "Opaque".

Examples of suitable formats for Transparent copies include plain ASCII without markup, Texinfo input format, LaTeX input format, SGML or XML using a publicly available DTD, and standard-conforming simple HTML, PostScript or PDF designed for human modification. Examples of transparent image formats include PNG, XCF and JPG. Opaque formats include proprietary formats that can be read and edited only by proprietary word processors, SGML or XML for which the DTD and/or processing tools are not generally available, and the machine-generated HTML, PostScript or PDF produced by some word processors for output purposes only.

The "Title Page" means, for a printed book, the title page itself, plus such following pages as are needed to hold, legibly, the material this License requires to appear in the title page. For works in formats which do not have any title page as such, "Title Page" means the text near the most prominent appearance of the work's title, preceding the beginning of the body of the text.

A section "Entitled XYZ" means a named subunit of the Document whose title either is precisely XYZ or contains XYZ in parentheses following text that translates XYZ in another language. (Here XYZ stands for a specific section name mentioned below, such as "Acknowledgements", "Dedications", "Endorsements", or "History".) To "Preserve the Title" of such a section when you modify the Document means that it remains a section "Entitled XYZ" according to this definition.

The Document may include Warranty Disclaimers next to the notice which states that this License applies to the Document. These Warranty Disclaimers are considered to be included by reference in this License, but only as regards disclaiming warranties: any other implication that these Warranty Disclaimers may have is void and has no effect on the meaning of this License.

You may copy and distribute the Document in any medium, either commercially or noncommercially, provided that this License, the copyright notices, and the license notice saying this License applies to the Document are reproduced in all copies, and that you add no other conditions whatsoever to those of this License. You may not use technical measures to obstruct or control the reading or further copying of the copies you make or distribute. However, you may accept compensation in exchange for copies. If you distribute a large enough number of copies you must also follow the conditions in section 3.

You may also lend copies, under the same conditions stated above, and you may publicly display copies.

If you publish printed copies (or copies in media that commonly have printed covers) of the Document, numbering more than 100, and the Document's license notice requires Cover Texts, you must enclose the copies in covers that carry, clearly and legibly, all these Cover Texts: Front-Cover Texts on the front cover, and Back-Cover Texts on the back cover. Both covers must also clearly and legibly identify you as the publisher of these copies. The front cover must present the full title with all words of the title equally prominent and visible. You may add other material on the covers in addition. Copying with changes limited to the covers, as long as they preserve the title of the Document and satisfy these conditions, can be treated as verbatim copying in other respects.

If the required texts for either cover are too voluminous to fit legibly, you should put the first ones listed (as many as fit reasonably) on the actual cover, and continue the rest onto adjacent pages.

If you publish or distribute Opaque copies of the Document numbering more than 100, you must either include a machine-readable Transparent copy along with each Opaque copy, or state in or with each Opaque copy a computer-network location from which the general network-using public has access to download using public-standard network protocols a complete Transparent copy of the Document, free of added material. If you use the latter option, you must take reasonably prudent steps, when you begin distribution of Opaque copies in quantity, to ensure that this Transparent copy will remain thus accessible at the stated location until at least one year after the last time you distribute an Opaque copy (directly or through your agents or retailers) of that edition to the public.

It is requested, but not required, that you contact the authors of the Document well before redistributing any large number of copies, to give them a chance to provide you with an updated version of the Document.

You may copy and distribute a Modified Version of the Document under the conditions of sections 2 and 3 above, provided that you release the Modified Version under precisely this License, with the Modified Version filling the role of the Document, thus licensing distribution and modification of the Modified Version to whoever possesses a copy of it. In addition, you must do these things in the Modified Version:

  1. Use in the Title Page (and on the covers, if any) a title distinct from that of the Document, and from those of previous versions (which should, if there were any, be listed in the History section of the Document). You may use the same title as a previous version if the original publisher of that version gives permission.

  2. List on the Title Page, as authors, one or more persons or entities responsible for authorship of the modifications in the Modified Version, together with at least five of the principal authors of the Document (all of its principal authors, if it has fewer than five), unless they release you from this requirement.

  3. State on the Title page the name of the publisher of the Modified Version, as the publisher.

  4. Preserve all the copyright notices of the Document.

  5. Add an appropriate copyright notice for your modifications adjacent to the other copyright notices.

  6. Include, immediately after the copyright notices, a license notice giving the public permission to use the Modified Version under the terms of this License, in the form shown in the Addendum below.

  7. Preserve in that license notice the full lists of Invariant Sections and required Cover Texts given in the Document's license notice.

  8. Include an unaltered copy of this License.

  9. Preserve the section Entitled "History", Preserve its Title, and add to it an item stating at least the title, year, new authors, and publisher of the Modified Version as given on the Title Page. If there is no section Entitled "History" in the Document, create one stating the title, year, authors, and publisher of the Document as given on its Title Page, then add an item describing the Modified Version as stated in the previous sentence.

  10. Preserve the network location, if any, given in the Document for public access to a Transparent copy of the Document, and likewise the network locations given in the Document for previous versions it was based on. These may be placed in the "History" section. You may omit a network location for a work that was published at least four years before the Document itself, or if the original publisher of the version it refers to gives permission.

  11. For any section Entitled "Acknowledgements" or "Dedications", Preserve the Title of the section, and preserve in the section all the substance and tone of each of the contributor acknowledgements and/or dedications given therein.

  12. Preserve all the Invariant Sections of the Document, unaltered in their text and in their titles. Section numbers or the equivalent are not considered part of the section titles.

  13. Delete any section Entitled "Endorsements". Such a section may not be included in the Modified Version.

  14. Do not retitle any existing section to be Entitled "Endorsements" or to conflict in title with any Invariant Section.

  15. Preserve any Warranty Disclaimers.

If the Modified Version includes new front-matter sections or appendices that qualify as Secondary Sections and contain no material copied from the Document, you may at your option designate some or all of these sections as invariant. To do this, add their titles to the list of Invariant Sections in the Modified Version's license notice. These titles must be distinct from any other section titles.

You may add a section Entitled "Endorsements", provided it contains nothing but endorsements of your Modified Version by various parties--for example, statements of peer review or that the text has been approved by an organization as the authoritative definition of a standard.

You may add a passage of up to five words as a Front-Cover Text, and a passage of up to 25 words as a Back-Cover Text, to the end of the list of Cover Texts in the Modified Version. Only one passage of Front-Cover Text and one of Back-Cover Text may be added by (or through arrangements made by) any one entity. If the Document already includes a cover text for the same cover, previously added by you or by arrangement made by the same entity you are acting on behalf of, you may not add another; but you may replace the old one, on explicit permission from the previous publisher that added the old one.

The author(s) and publisher(s) of the Document do not by this License give permission to use their names for publicity for or to assert or imply endorsement of any Modified Version.

You may combine the Document with other documents released under this License, under the terms defined in section 4 above for modified versions, provided that you include in the combination all of the Invariant Sections of all of the original documents, unmodified, and list them all as Invariant Sections of your combined work in its license notice, and that you preserve all their Warranty Disclaimers.

The combined work need only contain one copy of this License, and multiple identical Invariant Sections may be replaced with a single copy. If there are multiple Invariant Sections with the same name but different contents, make the title of each such section unique by adding at the end of it, in parentheses, the name of the original author or publisher of that section if known, or else a unique number. Make the same adjustment to the section titles in the list of Invariant Sections in the license notice of the combined work.

In the combination, you must combine any sections Entitled "History" in the various original documents, forming one section Entitled "History"; likewise combine any sections Entitled "Acknowledgements", and any sections Entitled "Dedications". You must delete all sections Entitled "Endorsements".

You may make a collection consisting of the Document and other documents released under this License, and replace the individual copies of this License in the various documents with a single copy that is included in the collection, provided that you follow the rules of this License for verbatim copying of each of the documents in all other respects.

You may extract a single document from such a collection, and distribute it individually under this License, provided you insert a copy of this License into the extracted document, and follow this License in all other respects regarding verbatim copying of that document.

A compilation of the Document or its derivatives with other separate and independent documents or works, in or on a volume of a storage or distribution medium, is called an "aggregate" if the copyright resulting from the compilation is not used to limit the legal rights of the compilation's users beyond what the individual works permit. When the Document is included in an aggregate, this License does not apply to the other works in the aggregate which are not themselves derivative works of the Document.

If the Cover Text requirement of section 3 is applicable to these copies of the Document, then if the Document is less than one half of the entire aggregate, the Document's Cover Texts may be placed on covers that bracket the Document within the aggregate, or the electronic equivalent of covers if the Document is in electronic form. Otherwise they must appear on printed covers that bracket the whole aggregate.

Translation is considered a kind of modification, so you may distribute translations of the Document under the terms of section 4. Replacing Invariant Sections with translations requires special permission from their copyright holders, but you may include translations of some or all Invariant Sections in addition to the original versions of these Invariant Sections. You may include a translation of this License, and all the license notices in the Document, and any Warranty Disclaimers, provided that you also include the original English version of this License and the original versions of those notices and disclaimers. In case of a disagreement between the translation and the original version of this License or a notice or disclaimer, the original version will prevail.

If a section in the Document is Entitled "Acknowledgements", "Dedications", or "History", the requirement (section 4) to Preserve its Title (section 1) will typically require changing the actual title.

You may not copy, modify, sublicense, or distribute the Document except as expressly provided for under this License. Any other attempt to copy, modify, sublicense or distribute the Document is void, and will automatically terminate your rights under this License. However, parties who have received copies, or rights, from you under this License will not have their licenses terminated so long as such parties remain in full compliance.

The Free Software Foundation may publish new, revised versions of the GNU Free Documentation License from time to time. Such new versions will be similar in spirit to the present version, but may differ in detail to address new problems or concerns. See http://www.gnu.org/copyleft/.

Each version of the License is given a distinguishing version number. If the Document specifies that a particular numbered version of this License "or any later version" applies to it, you have the option of following the terms and conditions either of that specified version or of any later version that has been published (not as a draft) by the Free Software Foundation. If the Document does not specify a version number of this License, you may choose any version ever published (not as a draft) by the Free Software Foundation.

Copyright (c) YEAR YOUR NAME.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.2
or any later version published by the Free Software Foundation;
with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts.
A copy of the license is included in the section entitled "GNU
 Free Documentation License".

If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts, replace the "with...Texts". line with this:

with the Invariant Sections being LIST THEIR TITLES, with the
Front-Cover Texts being LIST, and with the Back-Cover Texts being LIST.

If you have Invariant Sections without Cover Texts, or some other combination of the three, merge those two alternatives to suit the situation.

If your document contains nontrivial examples of program code, we recommend releasing these examples in parallel under your choice of free software license, such as the GNU General Public License, to permit their use in free software.