OpenVZ Forum: Devel » Re: [Vserver] Re: Container Test Campaign

Home » Mailing lists » Devel » Re: [Vserver] Re: Container Test Campaign

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: [Vserver] Re: Container Test Campaign [message #4277]

Tue, 04 July 2006 15:34

Cedric Le Goater
Messages: 443
Registered: February 2006

Senior Member

Kirill Korotaev wrote:
> Cedric,
>
>> these informations are not explicit yet but please check the raw data,
>> for
>> example :
>>
>> http://lxc.sourceforge.net/bench/r3/dbenchraw
>>
>> you will see that each test is run nearly 100 times. the 5% min and max
>> values are stripped before doing an average. min, max and std dev are
>> missing in the graph. soon to come. the raw data already contains some
>> interesting information on the results.
>>
>> cheers,
>
> can you please remake this test for OpenVZ with iosched=as?

ok, we will, with and without. It will add one bar to the graph.

thanks for the feedback,

C.

Report message to a moderator

Re: [Vserver] Re: Container Test Campaign [message #4343 is a reply to message #4277]

Thu, 06 July 2006 10:48

Herbert Poetzl
Messages: 239
Registered: February 2006

Senior Member

On Tue, Jul 04, 2006 at 05:34:23PM +0200, Cedric Le Goater wrote:
> Kirill Korotaev wrote:
> > Cedric,
> >
> >> these informations are not explicit yet but please check the raw data,
> >> for
> >> example :
> >>
> >> http://lxc.sourceforge.net/bench/r3/dbenchraw
> >>
> >> you will see that each test is run nearly 100 times. the 5% min and max
> >> values are stripped before doing an average. min, max and std dev are
> >> missing in the graph. soon to come. the raw data already contains some
> >> interesting information on the results.
> >>
> >> cheers,
> >
> > can you please remake this test for OpenVZ with iosched=as?
>
> ok, we will, with and without. It will add one bar to the graph.

I'd suggest to test _all_ available schedulers if possible,
for example Linux-VServer decided to favor the cfq scheduler
for 'fair' I/O scheduling per context, and OVZ did similar
(IIRC)

TIA,
Herbert

> thanks for the feedback,
>
> C.
> _______________________________________________
> Vserver mailing list
> Vserver@list.linux-vserver.org
> http://list.linux-vserver.org/mailman/listinfo/vserver

Report message to a moderator

Re: [Vserver] Re: Container Test Campaign [message #4346 is a reply to message #4343]

Thu, 06 July 2006 11:00

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

>>ok, we will, with and without. It will add one bar to the graph.
>
>
> I'd suggest to test _all_ available schedulers if possible,
> for example Linux-VServer decided to favor the cfq scheduler
> for 'fair' I/O scheduling per context, and OVZ did similar
> (IIRC)

For OpenVZ the reason is not CFQ fair scheduling only
(which is not that fair actually in our tests done inhouse),
but more in the fact that anticipatory scheduler has some bad issues
with sync. on some real life workloads (e.g. with SMTP agents)
anticipatory fails to work with reasonable performance.
http://bugzilla.kernel.org/show_bug.cgi?id=5900

Thanks,
Kirill

Report message to a moderator

Previous Topic:	patch against 2.6.8.1 (stable)
Next Topic:	Patch for vzctl-3.0.10.tar.bz2 on Debian Sarge systems

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Sat Sep 14 11:43:12 GMT 2024

Total time taken to generate the page: 0.04072 seconds