Distribute PCI Requests Across Multiple Devices
PCI requests are provisioned in a list based fashion. In SR-IOV networking devices, a set of candidate virtual functions can span multiple physical functions and physical ports. Distributing single and multiple device requests across multiple physical functions provides:
* Better load distribution across available links.
* Provides L2 redundancy when multiple devices are allocated on behalf of a single guest.
Blueprint information
- Status:
- Not started
- Approver:
- John Garbutt
- Priority:
- Low
- Drafter:
- Brent Eagles
- Direction:
- Needs approval
- Assignee:
- Roman Bogorodskiy
- Definition:
- Pending Approval
- Series goal:
- None
- Implementation:
-
Not started
- Milestone target:
-
next
- Started by
- Completed by
Related branches
Related bugs
Sprints
Whiteboard
Related spec:
https:/
Gerrit topic: https:/
Addressed by: https:/
Distribute PCI Requests Across Multiple Devices
Some questions that need to be answered for the spec:
- What are the implications with respect to NUMA?
- What is the expectation/
- Do we need configuration on this? What form should it take (flavor, config, etc)?
- Should distribution be the default behavior and should the inability to distribute result in a failure? (Actually, just asking this it would seem wrong to fail by default at the beginning because that breaks backwards compatibility - disable by default and figure out some deprecation cycle?)
- Implications on migration?
Pragmatic questions for planning and implementation:
- What cross-cutting efforts are underway that need to be considered (e.g. refactoring, moving to an oslo lib for something, does microversions come into play anywhere, etc.)
- Who can test it for real?
As there is no code up for review, removing this blueprint from mitaka, as part of the non-priority Feature Freeze. For more details please see: http://
