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

RE: [Xen-users] Poor performance running apache under xen


  • To: <Richard@xxxxxxxxxxxxxxx>, <xen-users@xxxxxxxxxxxxxxxxxxx>
  • From: "Adam Crews -X \(acrews - Xerox at Cisco\)" <acrews@xxxxxxxxx>
  • Date: Thu, 27 Jul 2006 10:38:03 -0700
  • Authentication-results: sj-dkim-3.cisco.com; header.From=acrews@xxxxxxxxx; dkim=pass ( sig from cisco.com verified; );
  • Delivery-date: Fri, 28 Jul 2006 02:49:19 -0700
  • List-id: Xen user discussion <xen-users.lists.xensource.com>
  • Thread-index: AcaxO070UohWx6KbQt66E1Pteg8ZggAZ/0+A
  • Thread-topic: [Xen-users] Poor performance running apache under xen

This sounds similar to a problem I had.  For my system setting
"EnableSendfile off" in the apache config fixed things for me.

-Adam

  | -----Original Message-----
  | From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
  | [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
  | Richard@xxxxxxxxxxxxxxx
  | Sent: Wednesday, July 26, 2006 10:12 PM
  | To: xen-users@xxxxxxxxxxxxxxxxxxx
  | Subject: [Xen-users] Poor performance running apache under xen
  | 
  | We've been seeing a fairly significant slow down of a 
  | fairly large, in
  | house java application based on apache tomcat, when running 
  | under xen.
  | In order to gain some insight into this I've run similar tests using
  | Apache2 and ApacheBench and we are seeing almost exactly 
  | the same slow
  | down of about 24% (23.8% for our application and 23.5% for apache).
  | 
  | The test consisted of sending one million reqests to the 
  | web server and
  | was run from a different machine over gigbit ethernet. 
  | During the tests
  | the driving machine did not go over 15% cpu load and the 1 
  | min load avg
  | did not go over 2. The machine under test had nothing else 
  | running. The
  | results given are for the second of two runs after a reboot. The cpu
  | usage is the average over the entire run obtained from sar 
  | with a four
  | second average (sar 4 0). The cpu figure is of the machine 
  | running the
  | test. The /lib/tls directory was removed.
  | 
  | The tests where carried out using the following versions of 
  | software:
  | 
  | * Ubuntu 5.10 (Breezy Badger) using Debian 32bit kernels
  | * Xen 3.0.2
  | * Apache 2.0.54 using both the mpm-prefork (multi-process 
  | worker) and the
  |   mpm-worker (hybrid multi-threaded multi-process worker)
  | * ApacheBench 2.0.41-dev
  | 
  | The different kernel configurations are:
  | 
  |     Stock kernel: 2.6.16-2-686-smp #1 SMP
  |     dom0 kernel: 2.6.16-1-xen-686 #1 SMP
  |     domU kernel: 2.6.16-1-xen-686 #1 SMP
  | 
  | The throughput is:
  | 
  |         Kernel                  throughput  % cpu (user + system)
  |     stock kernel (prefork):     5019.42     24.31
  |     stock kernel (worker):      4800.20     30.37
  |     dom0 kernel (prefork):      5029.60     22.37
  |     dom0 kernel (worker):       4843.59     29.97
  |     domU kernel (prefork):      3726.44     38.80
  |     domU kernel (worker):       3704.95     41.32
  | 
  | The reason I tested both the mpm-prefork and mpm-worker is 
  | test to see
  | if the threading concurrency model made a big difference -- 
  | it didn't!
  | 
  | Can anyone shed any light on these results? Is this the normal
  | performance that is to be expected from this type of application?
  | 
  | I've put all the results (ApacheBench results and the sar 
  | outputs) on
  | http://twosecondmemory.org/xen-performance/
  | 
  | rgh
  | 
  | 
  | _______________________________________________
  | Xen-users mailing list
  | Xen-users@xxxxxxxxxxxxxxxxxxx
  | http://lists.xensource.com/xen-users
  | 

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


 


Rackspace

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