docs/doc/source/guest_integration/openstack/pci-irq-affinity-4fed257d094b.rst
Elisamara Aoki Goncalves 12b6ff7cd4 Document PCI IRQ Affinity Agent creation (pick r6)
This commit includes the creation of a doc about how the agent operates
and how to configure OpenStack to operate properly with it.
Fixed minor editorial issues.

Signed-off-by: Elisamara Aoki Goncalves <elisamaraaoki.goncalves@windriver.com>
Change-Id: Ie118b1b2b7924a7f3a94bec089cd315094ebdef2
2022-04-13 17:47:37 -03:00

2.1 KiB

PCI IRQ Affinity

While in OpenStack it is possible to enable instances to use devices, the interrupts generated by these devices may be handled by host CPUs that are unrelated to the instance, and this can lead to noisy neighbor issues and lower performance, if the device interrupts were handled by the instance's CPUs.

Affinity can only be specified for instances with dedicated vCPUs.

Affinity assignments are used to achieve higher performance since the instance's core(s) will handle the interrupts from devices used by these instances. Interrupts are thus prevented from consuming cycles on the platform cores.

Flavor spec

A specific flavor spec for interrupt affining is used to determine which vCPUs assigned to the instance must handle the interrupts from the devices:

hw:pci_irq_affinity_mask=<vcpus_cpulist>

Where vcpus_cpulist can assume a comma-separated list of values that can be expressed as:

  • int: the vCPU expressed by int will be assigned to handle the interruptions from the devices
  • int1-int2: the vCPUs between int1 and int2 (inclusive) will be used to handle the interruptions from the devices
  • ^int: the vCPU expressed by int will not be assigned to handle the interruptions from the devices and shall be used to exclude a vCPU that was included in a previous range

Note

int must be a value between 0 and flavor.vcpus - 1.

For example: hw_pci_irq_affinity_mask=1-4,^3,6 means that vCPUs with indexes 1,2,4 and 6 from the vCPU list that Nova allocates to the instance will be assigned to handle interruptions from the devices.

Limitations

  • No CPU affining is supported for instances using shared CPUs (i.e., when using flavor spec hw:cpu_policy=shared).
  • No CPU affining will be performed when invalid ranges are specified in the flavor spec; a log error message will be generated indicating the problem.