[openstack-dev] cgroups cpu share allocation in grizzly seems incorrect

Qiu Yu unicell at gmail.com
Tue Aug 27 06:53:20 UTC 2013


On Fri, Aug 23, 2013 at 6:00 AM, Chris Friesen
<chris.friesen at windriver.com> wrote:
>
> I just noticed that in Grizzly regardless of the number of vCPUs the value
> of /sys/fs/cgroup/cpu/libvirt/qemu/instance-X/cpu.shares seems to be the
> same.  If we were overloaded, this would give all instances the same cpu
> time regardless of the number of vCPUs in the instance.
>
> Is this design intent?  It seems to me that it would be more correct to have
> the instance value be multiplied by the number of vCPUs.

I think it make sense to have each vCPU an equal weight for scheduling
sake. This makes each vCPU an equal entity with same computing power.

For limiting vCPU hard limit purpose, using CFS quota/period, please
check following BP for reference.
https://blueprints.launchpad.net/nova/+spec/quota-instance-resource

--
Qiu Yu



More information about the OpenStack-dev mailing list