-
Suggestion
-
Resolution: Done
-
P3: Somewhat important
-
None
-
None
-
None
Summary
Virtual machines have better performance if they are constrained to run on single NUMA node and currently they are not.
Current situation (mature-bear, 12th of March 2018)
ubuntu@mature-bear:~$ numactl --hardware available: 2 nodes (0-1) node 0 cpus: 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 node 0 size: 128843 MB node 0 free: 27940 MB node 1 cpus: 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 node 1 size: 129016 MB node 1 free: 30953 MB node distances: node 0 1 0: 10 21 1: 21 10
- There are two NUMA nodes
- You can extract the same information from <topoplogy> of "virsh capabilities"
CPU affinity of virtual machines:
ubuntu@mature-bear:~$ virsh vcpuinfo one-451747 setlocale: No such file or directory VCPU: 0 CPU: 74 State: running CPU time: 907.4s CPU Affinity: yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy VCPU: 1 CPU: 8 State: running CPU time: 839.2s CPU Affinity: yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy
- VM guest has 2 virtual CPUs which can run on any physical HW
- Doing "virsh cpu-stats one-XXXXX" confirms that guest VMs are running on all CPUs
Suggestion
It should be benchmarked if we get better guest VM performance by constraining the physical CPU allocation to NUMA nodes.
Links
Information taken from here:
https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/5/html/virtualization/ch33s08
http://www.admin-magazine.com/Archive/2014/20/Best-practices-for-KVM-on-NUMA-servers
Possible places how to configure NUMA for the guests:
- mentioned in
-
Page Loading...