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

Re: [Xen-devel] [PATCH] xen: Implement domain runstates


  • To: Keir Fraser <keir@xxxxxxx>
  • From: George Dunlap <George.Dunlap@xxxxxxxxxxxxx>
  • Date: Fri, 26 Nov 2010 11:52:52 +0000
  • Cc: "xen-devel@xxxxxxxxxxxxxxxxxxx" <xen-devel@xxxxxxxxxxxxxxxxxxx>, Bruce Edge <bruce.edge@xxxxxxxxx>
  • Delivery-date: Fri, 26 Nov 2010 03:53:54 -0800
  • Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:sender:in-reply-to:references:date :x-google-sender-auth:message-id:subject:from:to:cc:content-type; b=Wy4kdBK5k47WpLrkqeKgVqs6pAMy2H+93C8wUCWXusTRwBqMcxCE2fObmqDbd1ueGL 3vtNwNiv3n2ryp+YVmrHSfXpe2pnnWc6Mv7b8AfOS3Dj+Fo+UN2yX2jDIqIqUxETIidb QxMX/waiWCkiF0YkVfIYtBTlWW9hOSaTj+OM0=
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>

On Thu, Nov 25, 2010 at 10:24 PM, Keir Fraser <keir@xxxxxxx> wrote:
> You may be under the impression the feature is free. It isn't -- it adds a
> lock and loop over all vcpus on every vcpu scheduling/descheduling decision.
> That could be frequent for VCPUs doing a lot of short blocking, for example.
> And the system-wide work done to support this feature will scale
> quadratically on average with vcpus-per-domain. My gut feeling is this isn't
> a good thing to have on your context switch path.

That's certainly reasonable.  If I get a chance to do some perf
testing over the next few weeks, I'll post some performance numbers,
and we can quantify the slowdown (although catching potential cache
"cliffs" may be more difficult).

Would you consider accepting it as a feature off by default, enabled
by a Xen command-line parameter?

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

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