Sorry, I completely missed the response hitting my inbox. That this still shows the NUMA issues is a bit unexpected. So looking at the log I must have done something wrong with the RC printout of:
parse node returns (RC=-1) sockets=1 cores=6 threads=2 If that really was -1 it would not have iterated over the second node. The output is interesting. So the content of sockets, cores, and threads seems to match the info you gave in comment #29. From that cpuinfo there I assume the node0 directory contains links to cpu0-5 and cpu12-17. So the 6 cores with hyperthreading yielding 12 logical cpus on that node. Likewise node1 would have links for cpu6-11 and cpu18-23. Which all looks good and sensible. And from what is currently logged I can not see how the nodeinfo would get wrong. With one exception, but that would cause a different wrong info of node=1 sockets=1 cores=>24 and threads=1. This check is new in libvirt 1.2.12, so would not explain the previous wrong count. It basically checks whether nodes*sockets*cores*thread is the same as cpus+offline (should be 2*1*6*2 == 24+0 from the log). Can you let me know what nodeinfo actually printed? -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1446177 Title: Nodeinfo returns wrong NUMA topology / bad virtualization performance To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/libvirt/+bug/1446177/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs