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

[Xen-devel] Long system ticks in guest domain


  • To: <xen-devel@xxxxxxxxxxxxxxxxxxx>
  • From: "Donald MacGregor" <donald.macgregor@xxxxxxxxxxxx>
  • Date: Wed, 27 Jun 2007 13:31:26 -0400
  • Delivery-date: Wed, 27 Jun 2007 10:29:26 -0700
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>
  • Thread-index: Ace44P1drhymfNWhSYKnGjPqgxsIcA==
  • Thread-topic: Long system ticks in guest domain

I've been working on running a SIP call server on a Xen guest domain - this 
requires near realtime performance, so I've been testing to see how Xen handles 
it. So far it's been going pretty well, except I've noticed an odd behaviour 
under load. This software runs partially as a kernel object, and I've modified 
it to throw some debug messages out when it detects system timer updates (and 
the time since the last update was received in nanoseconds). Under light load 
it receives the updates every jiffy (so 4ms or 1ms depending on the HZ setting) 
.. under heavy load this update starts taking longer and longer, topping out at 
an update every 10ms. Basically my questions are, is this expected behaviour 
and what might cause it?

I've tried a few things to see if it changed the results, including giving the 
guest domain a dedicated cpu (taken if off dom0's list to use). I've also tried 
changing the scheduler to sedf and modified the slice and period to be small 
values. Other than running the guest with different jiffy periods (made no 
difference on the upper bound of the updates) those have been my primary 
attempts.

Any help on this would be much appreciated.

Thanks,
Don

_______________________________________________
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®.