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

[Xen-devel] SEDF unable to enforce hard shares under I/O intensive workloads?

  • To: xen-devel@xxxxxxxxxxxxxxxxxxx, Stephan Diestelhorst <sd386@xxxxxxxxx>
  • From: Diwaker Gupta <diwaker.lists@xxxxxxxxx>
  • Date: Sun, 8 Jan 2006 15:06:00 -0800
  • Delivery-date: Sun, 08 Jan 2006 23:12:12 +0000
  • Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:mime-version:content-type:content-transfer-encoding:content-disposition; b=bV5J5rWtpVGtwM11x19hChC1IZzyZ7lNKxywegexC1wgQHkBO8jZEuFFYGidWoWLk78s9gLk9xWTHNmJPbk3cyQ5y5SU9hYh4P+YNuXfkq+8j8BgXWPNWY7aPhsdsVaQdrm3NC6JkDwIymNjOQfizTw3fUluBX/rDAkLESHK6vU=
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>


I have the following setup:

o VM1 hosts a webserver serving 10KB files
o VM2 hosts a webserver serving 100KB files
o dom0, VM1, VM2 all run on the same CPU (I'm running with nosmp, noht flags)
o SEDF weights are 2, 2, 2 (equal weights to all) with the extra time
flag set to 0 for all (so ideally every domain should get ~32% CPU and
no more)

I now load the web servers on VM1 and VM2 using external httperf
clients. In my tests, I can routinely see dom0 averaging 37% and
sometimes going as high as 40%.

Has anyone seen this behavior where under I/O intensive workloads SEDF
fails to enforce hard shares? Any ideas on how/why this might happen
will be much appreciated.

BTW, it seems in SEDF only 95% of the CPU is available to the domains.
So the remaining 5% is reserved for ?

Web/Blog/Gallery: http://floatingsun.net
Xen-devel mailing list



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