trying to understand steal time with cpu pinning

Sean Mooney smooney at redhat.com
Thu Jul 11 12:03:28 UTC 2019


On Thu, 2019-07-11 at 06:15 +0000, Manuel Sopena Ballesteros wrote:
> Dear Openstack community,
> 
> Please correct me if I am wrong.
> 
> As far as I understand `steal time > 0` means that the hypervisor has replaced a vcpu with a different one on the
> physical cpu.
> Also, cpu pinning allocates a vcpu to a physical cpu permanently.
> 
> I have a vm setup with cpu pinning and numa affinity and realized, that cpu steal time is between 1% and 0%.
> 
> Why is that?
there are 2 ways that this can happen.
1.) you are not setting hw:emulator_thread_policy to move the qemu emulator threads to a different core.
2.) you have host system process or kernel threads that are stealing guest cpu time like vhost threads.

you can prevent 2 using systemd or take teh blunt hammer approch and use the kernel isolcpus parmater
but that generally should only be used for realtime systems.

openstack does not prevent other host process form running on the vcpu_pin_set, that is left to the
operator/installer/os to do.

> 
> Thank you very much
> NOTICE
> Please consider the environment before printing this email. This message and any attachments are intended for the
> addressee named and may contain legally privileged/confidential/copyright information. If you are not the intended
> recipient, you should not read, use, disclose, copy or distribute this communication. If you have received this
> message in error please notify us at once by return email and then delete both messages. We accept no liability for
> the distribution of viruses or similar in electronic communications. This notice should not be removed.
Technically disclaimer like ^ are considered poor mailing list etiquette and are not legally enforceable.
if you can disable it for the openstack mailing list that would be for the best. the discloser,copy and distribution
parts in partcalar can never be honoured by a publicly archived mailing list so this just adds noise to the list.




More information about the openstack-discuss mailing list