Customizing instance CPU pinning policies¶
Important
The functionality described below is currently only supported by the libvirt/KVM driver.
By default, instance vCPU processes are not assigned to any particular host CPU, instead, they float across host CPUs like any other process. This allows for features like overcommitting of CPUs. In heavily contended systems, this provides optimal system performance at the expense of performance and latency for individual instances.
Some workloads require real-time or near real-time behavior, which is not possible with the latency introduced by the default CPU policy. For such workloads, it is beneficial to control which host CPUs are bound to an instance’s vCPUs. This process is known as pinning. No instance with pinned CPUs can use the CPUs of another pinned instance, thus preventing resource contention between instances. To configure a flavor to use pinned vCPUs, a use a dedicated CPU policy. To force this, run:
$ openstack flavor set m1.large --property hw:cpu_policy=dedicated
Caution
Host aggregates should be used to separate pinned instances from unpinned instances as the latter will not respect the resourcing requirements of the former.
When running workloads on SMT hosts, it is important to be aware of the impact that thread siblings can have. Thread siblings share a number of components and contention on these components can impact performance. To configure how to use threads, a CPU thread policy should be specified. For workloads where sharing benefits performance, use thread siblings. To force this, run:
$ openstack flavor set m1.large \
--property hw:cpu_policy=dedicated \
--property hw:cpu_thread_policy=require
For other workloads where performance is impacted by contention for resources, use non-thread siblings or non-SMT hosts. To force this, run:
$ openstack flavor set m1.large \
--property hw:cpu_policy=dedicated \
--property hw:cpu_thread_policy=isolate
Finally, for workloads where performance is minimally impacted, use thread siblings if available. This is the default, but it can be set explicitly:
$ openstack flavor set m1.large \
--property hw:cpu_policy=dedicated \
--property hw:cpu_thread_policy=prefer
For more information about the syntax for hw:cpu_policy and hw:cpu_thread_policy, refer to the
Flavors guide.
Applications are frequently packaged as images. For applications that require real-time or near real-time behavior, configure image metadata to ensure created instances are always pinned regardless of flavor. To configure an image to use pinned vCPUs and avoid thread siblings, run:
$ openstack image set [IMAGE_ID] \
--property hw_cpu_policy=dedicated \
--property hw_cpu_thread_policy=isolate
Image metadata takes precedence over flavor extra specs. Thus, configuring competing policies causes an exception. By setting a shared policy through image metadata, administrators can prevent users configuring CPU policies in flavors and impacting resource utilization. To configure this policy, run:...
