vCPUs count over-booking

[Claudio Chinicz]

Hi,

I noted that my sum total of vCPUs is larger than the actual number of vCPUs.

Maybe Xen works at the minimum in order to keep sum total of vCPUs within limits?

One thing I tried was to disable VT-x and then HVMs (min and max vCPUs is the same) would not start.

Any ideas of how Xen manages vCPUs within the limits of the processor?

Regards

[unman]

Interesting question, Claudio.
As I understand it Xen allows you to create arbitrary number of vCPUs,
unrelated to the actual number of pCPUs you have. As you allocate more
vCPUs the system will start scheduling calls to the pCPUs and this may
impact performance.I don't think that disabling VT-x will impact this,
and as you have discovered it means that HVMs will not function.
To work around the limit you may find it useful to reduce the number
of vCPUs in most "ordinary" qubes to 1. You can also limit the number
that dom0 uses, with the `dom0-max-vCPUs` parameter at boot.
If you have processor intensive qubes, you can then try allocating
specific vCPUs to that qube - this is called "pinning". In Xen this can
be done with the `xl vcpu-pin` command . You can make a pin hard ,"MUST
use this CPU", or soft, "PREFERS to use this CPU"
By default Qubes makes all qubes hard/soft pin to all CPUs.
If you think that you are being impacted by the scheduler, you can try
pinning on processor intensive qubes and see if it helps. There used to
be a health warning saying that pinning caused as many problems as it
(may) solve, so try at your own risk.

unman

[Claudio Chinicz]

Hi unman, thank you for the detailed explanation. Regards