[Beowulf] Performance characterising a HPC application
deadline at clustermonkey.net
Sun Mar 25 07:06:16 PDT 2007
> On Sat, Mar 24, 2007 at 11:08:32AM -0400, Douglas Eadline wrote:
>> You really need two networks for GAMMA as GAMMA turns off TCP
>> while in use. Not to difficult with today's GigE-full nodes.
>> Just need extra an extra switch.
> I'm familiar with that requirement, yes.
>> > I just got a X2100 M2 machine in today. 4 GBit ports, though 2 of them
>> > nVidia.
>> I was able to get 17 usecond latency (netpipe TCP ) with the nvidia
>> links on these systems. Out of the box broadcom was about
>> 40 useconds (netpipe TCP).
> Vanilla netpipe, stock Linux kernel?
Nothing special was done. This is a Sun Discovery
cluster. It came wired to the nvidia port. I was curious
and did some tests. Using Warewulf and RHEL4 to build
> That's a very good result,
> given that GAMMA with Broadcoms is only slightly better.
Yes, that what I thought as well (ran netpipe/TCP several
times to check it)
> Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
> ICBM: 48.07100, 11.36820 http://www.ativel.com
> 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
More information about the Beowulf