[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH 0/2] Credit2: introduce per-vcpu hard and soft affinity


Thank you for the initial feedback!

> First of all, this is, overall, a really good submission. The only thing
> you need to add is a few references.
> For example, you already sent something in the past (an RFC, IIRC).
> Mention it in the cover letter (I personally also add a link to the
> thread).

Yes I will add references in the cover letter as you suggest.

> So, how did the testing went? Since you're not saying anything, I
> imagine that, at least functionally, everything works, at least as far
> as you can tell, isn't it so? :-)

I tested the hard affinity patch and I observed that the pinned VCPUs
always ran on their assigned PCPUs. I used vcpu-pin to change the
assignments which also worked. I tested the soft affinity patch with
workloads that I believed would cause the VPCUs to stay on their soft
PCPUs and only move away from them to hard affinity only in certain
circumstances (all soft busy, hard idle). I will add more specifics to
the cover letter.

> At least as far as soft-affinity is concerned, it should not be too
> difficult to collect some numbers, to show that it is bringing benefits,
> at least for some workloads.
> Do you think you'll be able to do something like that?


> When doing this same thing for credit1, here's what I did:
> http://lists.xen.org/archives/html/xen-devel/2013-02/msg00009.html
> https://blog.xenproject.org/2013/03/14/numa-aware-scheduling-development-report/

I will take a look at these and try to run additional workloads. If
you have any other ideas about workloads I should try, please let me

Thank you!
Justin Weaver

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.