Re: [Xen-devel] network performance drop heavily in xen 4.0 release

[Pasi Kärkkäinen <pasik@xxxxxx>]

On Fri, Apr 16, 2010 at 12:24:12PM +0800, yingbin wang wrote:
> sorry, I forgot to report the performance of xen4.0+kernel2.6.18.8.
> 
> xen4.0+kernel2.6.18.8:
> [ ID] Interval       Transfer     Bandwidth
> [  4]  0.0-16.3 sec  1.79 GBytes    941 Mbits/sec
> 
> This combination can only meet some of our needs.
> 
> the reason why we didn't use 2.6.18.8:
>   1. Our storage application is base on NBD, which will goto deadlock
> on 2.6.18.8 when nbd-client and nbd-server are deployed on the same
> server.
>   2. Hard disk frequently offline
> 

Do you have Xen credit scheduler weights properly configured? 
http://wiki.xensource.com/xenwiki/XenBestPractices

> so we want to upgrade to kernel2.6.31.
> we also try xen4.0+kernel2.6.31.12(gentoo-xen-kernel patch). the
> performance is acceptable, but blktap2(support VHD) not work.
> has anybody solve the problem in xen4.0+kernel2.6.31.13?
> 

You could try pvops dom0 kernel from xen/stable-2.6.32.x branch aswell:
http://wiki.xensource.com/xenwiki/XenParavirtOps

-- Pasi

> Cheers,
> wyb
> 
> 2010/4/16 Keir Fraser <keir.fraser@xxxxxxxxxxxxx>:
> > On 15/04/2010 17:39, "yingbin wang" <yingbin.wangyb@xxxxxxxxx> wrote:
> >
> >>     I report a Bug !!!  We have just upgraded to
> >> xen4.0+kernel2.6.31.13 recently.  however , fond that the network
> >> performance drop heavily in dom0  (nearly Reduced by 2/3 vs
> >> xen3.4.2+kernel2.6.18.8 ) .
> >
> > How does xen4.0+kernel2.6.18.8 perform? The regression is more likely in the
> > dom0 kernel than Xen itself. And you don't *have* to upgrade both.
> >
> >  -- Keir
> >
> >
> >
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel

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