WARNING - OLD ARCHIVES

This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
   
 
 
Xen 
 
Home Products Support Community News
 
   
 

xen-users

[Xen-users] benchmarks + disappointing cstate results

To: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-users] benchmarks + disappointing cstate results
From: Andreas Kinzler <ml@xxxxxx>
Date: Fri, 23 Jul 2010 14:25:37 +0200
Delivery-date: Fri, 23 Jul 2010 05:27:53 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxxx
List-help: <mailto:xen-users-request@lists.xensource.com?subject=help>
List-id: Xen user discussion <xen-users.lists.xensource.com>
List-post: <mailto:xen-users@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/mailman/listinfo/xen-users>, <mailto:xen-users-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/mailman/listinfo/xen-users>, <mailto:xen-users-request@lists.xensource.com?subject=unsubscribe>
Sender: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
Hello,

I do a lot of benchmarking on Xen/KVM and recently started a new set of 
benchmark runs with
special attention to cstates. Since modern Intel CPUs need C3 state for their 
turbo mode I played
with Xen option max_cstate=1 to check the effect. Normally you would expect a 
drop of performance
with max_cstate=1 since it blocks the necessary precondition of Intel Nehalem 
to enable turbo mode.
My tests show the opposite instead! Using max_cstate=1 has no effect or causes 
increases in
performance instead. The conclusion: Xen cannot use turbo mode and even worse: 
without max_cstate=1
you are penalized even more.

Regards Andreas

date: 22.07.2010
Supermicro X8SIL-F, BIOS 1.1, Intel Xeon X3450, 8 GB RAM reg ECC
benchmark: time emerge apache (apache 2.2.15 using Gentoo on x86_64)
xen with parameters "dom0_mem=1024m dom0_max_vcpus=2"
pvops0 kernel from 14.07.10 (Jeremy's xen/stable-2.6.32.x)
xenified kernel 2.6.31-xen0-r14 (from Andrew Lyon)
xenified kernel 2.6.32-xen0-r3 (from Suse backported minimal)

native linux 2.6.31.12 with maxcpus=2
        real    0m41.193s
        user    0m41.479s
        sys     0m9.609s

2.6.32-xen0-r3 + xen 3.4.3, max_cstate=1
2.6.32-xen0-r3 + xen 3.4.2
2.6.32-xen0-r3 + xen-4.0.1_rc4, max_cstate=1
        real    0m53.122s
        user    0m41.795s
        sys     0m16.233s

2.6.31-xen0-r14 + xen 3.4.2
2.6.31-xen0-r14 + xen 3.4.2, max_cstate=1
        real    0m54.670s
        user    0m41.695s
        sys     0m18.757s

2.6.31-xen0-r14 + xen 3.4.3, max_cstate=1
        real    0m56.983s
        user    0m40.767s
        sys     0m20.597s

2.6.32.16-pvops0 + xen-4.0.1_rc4, max_cstate=1
2.6.32.16-pvops0 + xen-4.0.1_rc4
2.6.32.16-pvops0 + xen 3.4.3
        real    0m58.298s
        user    0m45.319s
        sys     0m30.126s

2.6.32-xen0-r3 + xen-4.0.1-rc4
        real    0m59.352s
        user    0m39.258s
        sys     0m20.405s

2.6.32-xen0-r3 + xen 3.4.3
        real    1m1.352s
        user    0m39.682s
        sys     0m21.645s

2.6.31-xen0-r14 + xen-4.0.1-rc4
2.6.31-xen0-r14 + xen 3.4.3
        real    1m9.633s
        user    0m41.735s
        sys     0m26.326s


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

<Prev in Thread] Current Thread [Next in Thread>
  • [Xen-users] benchmarks + disappointing cstate results, Andreas Kinzler <=