Uploaded image for project: 'Red Hat OpenStack Services on OpenShift'
  1. Red Hat OpenStack Services on OpenShift
  2. OSPRH-18

CPU state management with libvirt

XMLWordPrintable

    • CPU state management with libvirt
    • 5
    • False
    • Hide

      None

      Show
      None
    • False
    • OSPRH-120Compute Engineering Backlog
    • Committed
    • Committed
    • To Do
    • OSPRH-120 - Compute Engineering Backlog
    • openstack-nova-27.1.1-18.0.20230930093334.a869ab1.el9ost
    • Committed
    • Committed
    • 0% To Do, 8% In Progress, 92% Done
    • Hide
      .Manage dedicated CPU power state

      You can now configure the `nova-compute` service to manage dedicated CPU power state by setting [libvirt]cpu_power_management to True.

      This feature requires the Compute service to be set with [compute]cpu_dedicated_set. With that setting, all dedicated CPUs are powered down until they are used by an instance. They are powered up when an instance using them is booted. If power management is configured but [compute]cpu_dedicated_set isn't set, then the compute service will not start.

      By default, the power strategy offlines CPUs when powering down and onlines the CPUs on powering up, but another strategy is possible. Set [libvirt]cpu_power_management_strategy=governor to instead use governors, and use [libvirt]cpu_power_governor_low [libvirt]cpu_power_governor_high to direct which governors to use in online and offline mode (performance and powersave).
      Show
      .Manage dedicated CPU power state You can now configure the `nova-compute` service to manage dedicated CPU power state by setting [libvirt]cpu_power_management to True. This feature requires the Compute service to be set with [compute]cpu_dedicated_set. With that setting, all dedicated CPUs are powered down until they are used by an instance. They are powered up when an instance using them is booted. If power management is configured but [compute]cpu_dedicated_set isn't set, then the compute service will not start. By default, the power strategy offlines CPUs when powering down and onlines the CPUs on powering up, but another strategy is possible. Set [libvirt]cpu_power_management_strategy=governor to instead use governors, and use [libvirt]cpu_power_governor_low [libvirt]cpu_power_governor_high to direct which governors to use in online and offline mode (performance and powersave).
    • Enhancement
    • Done
    • Automated
    • OSPPlanningCycle3, 2023Q1
    • Compute
    • Approved
    • Red Hat OpenStack Services on OpenShift

      Problem statement:
      Customers running large openstack cloud with semi realtime long lived workloads
      like a telco cloud often have a workflow that involves packing host with VNFs
      that run for years at a time once they are deployed. while they take effort in
      the deployment order and vm tuning to ensure that as many vm are place on each host
      as reasonable they are faced with a tetris/bin packing problem where some cores
      are never used.
      (generally these vms are cpu pinned and ther is no memory or cpu over subscription)
      To that end as an operator of an OSP cloud i would like to be able to offline or power
      managed the unused cpu cores once my node is "full". it should also be simple to also
      revert this change if the node need to be used more dynamicaly again.

      the initial poc was completed in https://github.com/SeanMooney/arbiterd this epic tracks evolving that prototype to a project that can be upstreamed to opendev and proctised in osp.

            sbauza@redhat.com Sylvain Bauza
            smooney@redhat.com Sean Mooney
            rhos-dfg-compute
            Votes:
            1 Vote for this issue
            Watchers:
            13 Start watching this issue

              Created:
              Updated:
              Resolved: