Re: [Xen-users] The death of XEN by Novell

[Mark Williamson <mark.williamson@xxxxxxxxxxxx>]

On Sunday 20 July 2008, jim burns wrote:
> On Sun July 20 2008 9:51:52 am Venefax wrote:
> > The Novell engineer showed me a technical document, which I did not write
> > the number, on the Novell web site, where it said: windows SMP using the
> > Novell driver crashes.
>
> That's what I thought - the problem is with Novell's PV drivers, not Xen.
> James had some initial problems with SMP also.

Right, that's important.  Novell's PV drivers are their code so those drivers 
might have limitations that plain Xen does not.  Presumably it would be 
possible to run using fully emulated IO instead of PV drivers but I imagine 
this could have performance problems in your configuration.  Could be worth a 
try, if you can afford to experiment.

If the Windows VMs themselves are crashing due to Novell's PV drivers then I'm 
afraid this limitation / problem is really Novell's responsibility and they 
would be the best people to approach for assistance / complaints.  The code 
for those drivers isn't open, so only Novell can fix them.  If the Novell PV 
drivers are somehow crashing your Linux domains or dom0 or Xen itself then 
that might be a Xen bug and it would be useful to post more details here.

For comparison, Citrix/XenSource supply with their product PV drivers for 
Windows that are - AFAIK - a different codebase and may not have these 
limitations.  So do Virtual Iron, for their Xen-based product.  Both of these 
can only be used with their respective commercial hypervisors, AFAIK.  James 
Harper's PV drivers for Windows are GPL but still under development and so I 
doubt they'd be recommended for heavy production use.

There are a lot of different PV driver solutions for Windows about and none of 
them are officially part of the OSS Xen project; only one of them is even 
Open Source.

> >  I use a SIP Server on the Linux side, talking hundreds of times per
> > second to SQL Server, so I need SMP 8 ways on both  ends.
>
> Something I've never seen you actually say is that you *tried* vcpu=1, and
> the performance was bad. So far, it sounds like theory to me. Even with 1
> vcpu, playing with scheduling weights should help.

Venefax, are you able to try this?

Thank you,
Mark


-- 
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)

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