Depends on what you're asking.
The benefit of running two times two processes
per node probably stems from the fact that
applications usage of interconnect is temporal in
time. Hence, by running two application instances
per node, the interconnect usage is more spread in time.
The benefit of different policies of the
process-to-processor binding depends on more
issues. YMMV application wise. Further, some
multicore processor have a shared cache between
the cores (Woodcrest). Communication intensive
apps might then benefit from this cache. Other
apps tend to like running on two Opterons to get
the luxery of (mostly) using two memory controllers. And others don't.
Håkon
At 23:15 05.10.2006, Mark Hahn wrote:
On two-socket, dual core Opteron systems, we
often see that throughput is enhanced by using
two times two processes per node.
just for the obvious reason? (memory bandwidth)
--
Håkon Bugge
CTO
dir. +47 22 62 89 72
mob. +47 92 48 45 14
fax. +47 22 62 89 51
[EMAIL PROTECTED]
Skype: hakon_bugge
Scali - http://www.scali.com
Scaling the Linux Datacenter
_______________________________________________
Beowulf mailing list, [email protected]
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
