[Qemu-discuss] NUMA topology configuration question

[George Diamantopoulos]

Hello all,

I've been reading qemu documentation and so far it's unclear to me what is
expected to use in the "-numa" option and the "cpus" parameter in
particular:

* -numa node[,mem=size][,cpus=cpu[-cpu]][,nodeid=node]

Do the cpu IDs supplied here correspond to guest vCPUs, or to host pCPUs?

My guess is that is should be the former, but the only example I have been
able to find online alludes to the latter. You need to examine the lstopo
output the user supplies in their blog post and the recommended numa
options (below the libvirt XML configuration) to understand my confusion:

* https://tripleback.net/post/chasingvfioperformance/


My understanding is that when using NUMA topologies alongside CPU pinning,
one aims to do the following:

* pin vCPUs to pCPUs

* assign vCPUs to vNUMA topology nodes

* it is up to the user to pin vCPUs to pCPUs in such a way that all vCPUs
configured within a given vNUMA node are pinned to pCPUs belonging to the
same pNUMA node

The article/blog above seems to do the following instead:

* pin vCPUs to pCPUs

* assign pCPU IDs to vNUMA topology nodes in the example configuration
line, in a way * that pCPU IDs listed in in the same vNUMA node
configuration reside within the same pNUMA node


This doesn't make much sense to me. As qemu-doc suggests, -numa is a
resource assignment option, not a resource allocation one, so it is guest
resources that should appear here, not host resources.

Can somebody please confirm? Thanks!

Best regards,
George