On 6/10/20 9:36 AM, Patrick O'Callaghan wrote:
On 6/10/20 2:56 AM, Patrick O'Callaghan wrote:
In KVM/QEMU you can also pin specific cores to your VM to prevent
competition between the host and guest (mainly by avoiding cache
pollution as I understand it).
https://vfio.blogspot.com/2015/05/vfio-gpu-how-to-series-part-4-our-first.html
(search for vcpupin partway down the page). This doesn't relate to NUMA
systems but to cache efficiency (also mentioned in your second
reference), though it may be a matter of terminology. Perhaps I should
have said "avoid" or "mitigate" rather than "prevent".
I don't see anything there that suggests that CPU affinity is related to
competition between the host and the guest.
Setting CPU affinity correctly does improve cache use, and reduces the
frequency of a guest running on a CPU with a different path to memory
pages on NUMA systems, but scheduling the guest still requires that the
CPUs are available for scheduling. As guests are assigned more cores
(both when they're assigned all or nearly all of the system cores, or
when multiple guests are assigned a number of cores collectively greater
than the system total), you'll often see longer times between the guest
being scheduled to run (that is, greater CPU steal time). I still don't
know if that's the issue that is the subject of this thread, but that's
my suspicion.
_______________________________________________
users mailing list -- users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to users-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/users@xxxxxxxxxxxxxxxxxxxxxxx
