RE: [Xen-users] Large server, Xen limitations

[Steve Thompson <smt@xxxxxxxxxxxx>]

On Fri, 2 Jan 2009, Robert Dunkley wrote:

> Do you have a link or details of an example for the cygwin compiles you
> use as a sort of benchmark?
Unfortunately not. I did my initial testing when I installed the Dom0 last 
July. I didn't keep records as it was a single-developer setup (me), 
although of course I recorded all my measurements at the time.
I am compiling a package (of my own design) that is comprised of about 
500K lines of C and C++ code (mostly C), with about 1200 separate source 
and header files and about 100 makefiles. It is a fully native application 
on Windows, but I use cygwin with gcc/g++ -mno-cygwin and GNU make since 
the same makefiles can then be used on Windows as on the other platforms 
that are supported (RHEL3/4/5, CentOS 3/4/5, Fedora, Tru64, OSX). On Linux 
guests, the sources come from an NFS-mounted volume, and the objects go to 
a (different) NFS-mounted volume. On Windows, the same sources come from a 
Samba share, and the objects go to a (different) Samba share. Although I 
have 33 guests at present, I do the build on only 14 of them; the others 
are for something else. The NFS server and Samba server are on the Dom0 
(CentOS 5.2 x86_64, Dell PE2900 with 8 cores and 24 GB).
I found that 64-bit HVM guests were about 30% faster than 32-bit guests 
(where "faster" in this case relates to the elapsed time of a full build). 
In their current NFS-root configuration, 32-bit guests are a little faster 
than 64-bit, and 64-bit guests are about 50% faster than the same guests 
in HVM form. Performance of HVM Windows guests was dreadful until I 
installed PV drivers.
I still have four fully-HVM guests (RHEL3, RHEL4, 32 and 64 bit). I'll do 
some more timings and will report back.
Steve

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