Re: Dom0 cpu utilisation irregularly distributed

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 10/05/2011 07:22 AM, Fantu wrote:
> On some production servers with dom0 kernel 2.6.32 pvops and xen 4.0 and
> domus mainly windows with gplpv I noticed a strange cpu usage by the dom0
> mainly with the use of a single cpu, for example this take from one server:
> xm vcpu-list
> Name                                ID  VCPU   CPU State   Time(s) CPU
> Affinity
> ...
> Domain-0                             0     0     3   -b-  1043377.5 any cpu
> Domain-0                             0     1     2   r--  122083.9 any cpu
> Domain-0                             0     2     0   -b-  284636.1 any cpu
> Domain-0                             0     3     1   -b-   80772.9 any cpu
> ...
> Note the increased utilisation on the vcpu 0
>
> Domus cpu utilisation apparently successfully distributed but the dom0 has
> always use more of its vcpu 0
> Is this correct or there are bugs on kernel and/or xen?

VCPU use is entirely up to the guest kernels.  I suspect that cpu0 gets
heavier loading because there are a number of things which end up
running on cpu 0 by default, such as some kernel threads and interrupts.

You can try playing with interrupt migration on dom0 if you like.

But balanced CPU use isn't necessarily a good thing.  The kernel will
generally avoid migrating processes between CPUs needlessly, so unless
you're actually starved for CPU time, I would expect some VCPUs to be
underused compared to others.  It probably just means you've configured
dom0 with too many VCPUs.

    J