RE: ´ð¸´: [Xen-ia64-devel] Re: [PATCH 0/12] various fixes related the xenrelocation

[Alex Williamson <alex.williamson@xxxxxx>]

On Tue, 2008-01-08 at 15:46 +0100, tgingold@xxxxxxx wrote:
> Quoting "de Dinechin, Christophe (Integrity VM)"
> > Do you have rough performance benchmarks for Xen running a VT-i guest right
> > now? We discarded trap-and-emulate very early in the HPVM design because an
> > early experiment had shown that on Itanium, fault handling alone amounted to
> > something like 15% performance degradation. This may have changed with more
> > recent chips, but I'm curious about this.
> 
> I don't have figure but maybe some were sent in the mailing list.  For sure
> VT-i degradation is severe.  The Xen approach is para-virtualization where a
> tiny part of the OS is rewritten.  This approach is very effective but works
> only for open source OSes.

   Of course it depends on the workload.  I just ran some kernel build
tests last Friday on the latest bits and see a bit over 9% hit for a PV
Linux guest and only about 3% more for an HVM Linux guest.  That can be
reduced to about 2% over the PV case using PV-on-HVM block drivers.
These were for UP guests on a Montvale system with warm caches, so it's
mainly a memory and computation test.  YMMV with more I/O intensive
loads.  If anyone else has some performance numbers, please share.
Thanks,

        Alex
-- 
Alex Williamson                             HP Open Source & Linux Org.


_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel