[PATCH v2 2/9] drm/i915/gvt: create a dummy vGPU
Tian, Kevin
kevin.tian at intel.com
Fri Feb 17 09:48:14 UTC 2017
> From: Gao, Ping A
> Sent: Wednesday, February 15, 2017 10:04 AM
>
> On 2017/2/14 16:49, Tian, Kevin wrote:
> >> From: Ping Gao
> >> Sent: Tuesday, February 14, 2017 12:26 PM
> >>
> >> vGPU execution time is given by scheduler, it decide the resource
> >> vGPU could take. Creating a dummy vGPU mainly used to help time
> >> slice usage calculation and work as a time slice pool to consume
> >> the time slice that real vGPU no need. Besides that it also helpful
> >> to handle some special case.
> > do we need it even for weight-based scheduling?
>
> Yes, actually weight is another kind of timeslice limitation, e.g. two
> vGPU set weight as 5:2, that means the first vGPUs's max timeslice
> cannot exceed 5/(5+2); if both of them become idle, one of them has be
> chose as render owner, the time slice usage calculation of render owner
> is always go, when idle render owner has workload, the word load cannot
> execute timely as the timeslice maybe run out by it's idle status, it's
> has to wait for the next second to start, that's a problem here. Beside
> that introduce idle vGPU make framework more clearer and the calculation
> more easier.
yes, it makes sense.
More information about the intel-gvt-dev
mailing list