Description of problem: today if you have a multi numa node guest then all pci devices are assocated with virtual guest numa node0. this if the sriov device is associated with the virtual numa node 1 on the host this lead to cross qpi traffic forcine all acess to the device to go though virtual numa node 0. to prevent this nova should create at leat 1 pci root complex per virual guest numa node and map the pci device to the numa local pci root complex. Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
Starting with the Antelope upstream release, DFG:Compute is tracking its feature backlog in Jira. This feature request has been migrated to DFG:Compute’s new Jira backlog. The Jira tracker for this feature request can be found below. At the time of this writing, only Red Hat associates have access to the Jira tracker, but we are actively working on changing that as more and more of the OSP organization uses Jira for more things. https://issues.redhat.com/browse/OSP-1589