27 Virtualization: configuration options and settings #

A paravirtual machine can have up to 100 block devices composed of virtual CD readers and virtual disks. On paravirtual machines, virtual CD readers present the CD as a virtual disk with read-only access. Virtual CD readers cannot be used to write data to a CD.

After you have finished accessing a CD on a paravirtual machine, it is recommended that you remove the virtual CD reader from the virtual machine.

Paravirtualized guests can use the device type devtype=cdrom. This partly emulates the behavior of a real CD reader, and allows CDs to be changed. It is even possible to use the eject command to open the tray of the CD reader.

A fully virtual machine can have up to four block devices composed of virtual CD readers and virtual disks. A virtual CD reader on a fully virtual machine interacts with an inserted CD in the way you would expect a physical CD reader to interact.

When a CD is inserted in the physical CD reader on the host computer, all virtual machines with virtual CD readers based on the physical CD reader, such as /dev/cdrom/, can read the inserted CD. Assuming the operating system has automount functionality, the CD should automatically appear in the file system. Virtual CD readers cannot be used to write data to a CD. They are configured as read-only devices.

Virtual CD readers can be based on a CD inserted into the CD reader or on an ISO image file.

Although the default behavior of VNC viewer is to assign the first available port number, you should assign a specific VNC viewer port number to a specific virtual machine.

To assign a specific port number on a VM Guest, edit the xl setting of the virtual machine and change the vnclisten to the desired value. Note that for example for port number 5902, specify 2 only, as 5900 is added automatically:

vfb = [ 'vnc=1,vnclisten="localhost:2"' ]

For more information regarding editing the xl settings of a guest domain, see Section 25.1, “XL—Xen management tool”.

Tip

Assign higher port numbers to avoid conflict with port numbers assigned by the VNC viewer, which uses the lowest available port number.

If you access a virtual machine's display from the virtual machine host console (known as local or on-box access), you should use SDL instead of VNC viewer. VNC viewer is faster for viewing desktops over a network, but SDL is faster for viewing desktops from the same computer.

To set the default to use SDL instead of VNC, change the virtual machine's configuration information to the following. For instructions, see Section 25.1, “XL—Xen management tool”.

vfb = [ 'sdl=1' ]

Remember that, unlike a VNC viewer window, closing an SDL window terminates the virtual machine.

Dedicating CPU resources to Dom0 results in a better overall performance of the virtualized environment because Dom0 has free CPU time to process I/O requests from VM Guests. Failing to dedicate exclusive CPU resources to Dom0 usually results in a poor performance and can cause the VM Guests to function incorrectly.

Dedicating CPU resources involves three basic steps: modifying Xen boot line, binding Dom0's VCPUs to a physical processor, and configuring CPU-related options on VM Guests:

It is often necessary to dedicate specific CPU resources to a virtual machine. By default, a virtual machine uses any available CPU core. Its performance can be improved by assigning a reasonable number of physical processors to it as other VM Guests are not allowed to use them after that. Assuming a machine with 8 CPU cores while a virtual machine needs to use 2 of them, change its configuration file as follows:

vcpus=2
cpus="2,3"

The above example dedicates 2 processors to the VM Guest, and these being the 3rd and 4th one, (2 and 3 counted from zero). If you need to assign more physical processors, use the cpus="2-8" syntax.

If you need to change the CPU assignment for a guest named “alice” in a hotplug manner, do the following on the related Dom0:

> sudo xl vcpu-set alice 2
> sudo xl vcpu-pin alice 0 2
> sudo xl vcpu-pin alice 1 3

The example will dedicate 2 physical processors to the guest, and bind its VCPU 0 to physical processor 2 and VCPU 1 to physical processor 3. Now check the assignment:

> sudo xl vcpu-list alice
Name                              ID VCPUs   CPU State   Time(s) CPU Affinity
alice                             4     0     2   -b-       1.9 2-3
alice                             4     1     3   -b-       2.8 2-3

Just as with physical hardware, it is sometimes desirable to boot a VM Guest from a different device than its own boot device. For fully virtual machines, it is possible to select a boot device with the boot parameter in a domain xl configuration file:

boot = BOOT_DEVICE

BOOT_DEVICE can be one of c for hard disk, d for CD-ROM, or n for Network/PXE. You can specify multiple options, and they will be attempted in the given order. For example,

boot = dc

boots from CD-ROM, and falls back to the hard disk if CD-ROM is not bootable.

To be able to migrate a VM Guest from one VM Host Server to a different VM Host Server, the VM Guest system may only use CPU features that are available on both VM Host Server systems. If the actual CPUs are different on both hosts, it may be necessary to hide some features before the VM Guest is started. This maintains the possibility to migrate the VM Guest between both hosts. For fully virtualized guests, this can be achieved by configuring the cpuid that is available to the guest.

To gain an overview of the current CPU, have a look at /proc/cpuinfo. This contains all the important information that defines the current CPU.

To redefine a CPU, first have a look at the respective cpuid definitions of the CPU vendor. These are available from:

cpuid = "host,tm=0,sse3=0"

The syntax is a comma-separated list of key=value pairs, preceded by the word "host". A few keys take a numerical value, while all others take a single character which describes what to do with the feature bit. See man 5 xl.cfg for a complete list of cpuid keys. The respective bits may be changed by using the following values:

Note that counting bits is done from right to left, starting with bit 0.

In case you need to increase the default number of PCI-IRQs available to Dom0 and/or VM Guest, you can do so by modifying the Xen kernel command line. Use the command extra_guest_irqs= DOMU_IRGS,DOM0_IRGS. The optional first number DOMU_IRGS is common for all VM Guests, while the optional second number DOM0_IRGS (preceded by a comma) is for Dom0. Changing the setting for VM Guest has no impact on Dom0 and vice versa. For example to change Dom0 without changing VM Guest, use

extra_guest_irqs=,512