On 2016/10/19 22:37, Delsol, Thomas wrote:
I have the following questions regarding igvt-g solution :
1. Is there any documentation describing the igvt-g kernel boot
parameters (event_based_qos, preemption_policy,
The basic parameters were described in setup_guide.txt, the others you
mentioned are mostly for debug/experiment purpose, some of them are
obsolete like event_based_qos. I think it's good idea for us to summary
all of them together in the next release.
2. Is this possible to specify GPU priority per VM ?
If yes, how can it be achieved and is it possible to also set per
process priority within a given VM?
Currently It does not support GPU priority per VM, however it's in the
The minimum granularity for scheduler is VM, cannot control the process
in given VM. Do you have any special requirement about this?
3. Is this possible to guarantee GPU bandwidth for a particular VM and
if yes how can it be achieved ?
This Q3 release will support to set cap for each guest VM.
Cap control the maximum amount of vGPU resource will be able to consume,
valid value is [0, 100]. 0 and 100 means no upper cap. You can enable it
by set vgt_cap in QEMU config
iGVT-g mailing list