Train Cycle Themes

These are cycle themes. That means we intend to put particular effort into these initiatives. It does not mean we are committing to finishing any/all individual items in particular. It also does not mean these are the only things we will be concentrating on. Consider yourself disclaimed.

  1. Improve/expand Nova’s scheduling efficiency and scalability for all deployments, with particular focus on large-scale deployments. This means being able to ask Placement better questions so that more filtering is done in Placement to reduce the list of allocation candidates and minimize Nova-side filtering. Specifically we intend to target efforts such as:

    • NUMA structure modeling and affinity

    • Improved tracking of shared and dedicated logical processors

    • Reporting, tracking, and requesting additional resources on hosts

    • Supporting server group affinity and anti-affinity

    • Trait filters for driver capabilities, image types, and more

    • Forbidden trait and aggregate filters to isolate “special” hosts, avoid disabled nodes, etc.


    Much of the above work has dependencies on efforts in Placement. The Placement team’s cycle priorities are aligned accordingly.

  2. Enable requesting an instance with one or more accelerators either preprogrammed or dynamically programmed. This encompasses FPGAs managed by Cyborg as well as VGPUs (of multiple types) managed by Nova.


    This includes cross-project work with Cyborg. The Cyborg team’s cycle priorities are aligned accordingly.

  3. We want our documentation to be valid, easily referenced and generally suitable for purpose. We’re building on a strong foundation. Three objectives:

    1. Docs should be cleanly aligned to the directory structure. This is so end users can go to ‘/user’ and find the info they want without admin’y stuff thrown in. Ditto for admins, devs, etc.

    2. The install guide should work.

    3. Docs in the user and admin guides should be topic-focused and self-contained. Like we did with the console docs, a user/admin should be able to search Google for e.g. “attaching a PCI device” and find the guide that details what it is, how to enable it and how to use it. See for a non-OpenStack variant.