Re: [Xen-users] Xen 4.6 and Intel MKL Linpack behaviour

[Marko Đukić <marko.djukic@xxxxxxxxx>]

On 9 May 2016 at 17:32, Dario Faggioli <dario.faggioli@xxxxxxxxxx> wrote:
> On Mon, 2016-05-09 at 15:37 +0200, Marko Đukić wrote:
>> On 9 May 2016 at 14:57, Dario Faggioli <dario.faggioli@xxxxxxxxxx>
>> As you can see, in HVM guest shows 4 cpus and each entry shows "cpu
>> cores: 4" and a different core id for each entry.
>> In PVH guest shows "cpu cores: 1" and "core_id: 0" for each entry.
>>
> Ok, so you do get 4 vcpus, but they are all listed as if the were part
> of the same core.
>
> This happens because of the difference between CPUID (and topology
> handling in general) implementation between PV(H) and HVM that Roger
> was mentioning.
>
> There is work ongoing for improving this (making things more
> configurable and reliable), both on the Xen and Linux kernel sides, but
> patches are not upstream yet.
>
> If LINPACK relies on this 'core id' (or something equivalent/based on
> the same low level topology info), there's few that we can do, apart
> from changing LINPACK itself.
>
> Maybe (before embarking in changing LINPAK), fiddling with the cpuid
> config option is worth a try.
>
> For example, in this thread:
>  http://lists.xenproject.org/archives/html/xen-devel/2015-07/msg04423.html
>
> it's suggested to set it to:
>  cpuid=['0x1:ebx=xxxxxxxx00000001xxxxxxxxxxxxxxxx']
>
> This have an effect, and if it is not exactly what you want, you can
> change things a bit by playing with the value.
>
> This is a workaroud, and I don't know if it can be considered good
> enough for what you're tring to achieve.
>
> Regards,
> Dario
> --
> <<This happens because I choose it to happen!>> (Raistlin Majere)
> -----------------------------------------------------------------
> Dario Faggioli, Ph.D, http://about.me/dario.faggioli
> Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
>

Thank you very much, Dario, cpuinfo now shows 4 cores and LINPACK also
detects 4 cores. The GFlops LINPACK calculates are similar to the ones
in HVM guest, so apparently it works.

Regards

Marko

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users