[Beowulf] recommendations for cluster upgrades

Rahul Nabar rpnabar at gmail.com
Wed May 13 08:21:06 PDT 2009


On Wed, May 13, 2009 at 1:55 AM, Jan Heichler <jan.heichler at gmx.net> wrote:
> I'm really surprised that everyone just screams "Nehalem" - of course the
> platorm is the fastest that you can buy for money at the moment. It is the
> youngest design so that is not suprising.
>
> But clustering is always about price/performance. And AMD doesn't look so
> bad there.

Absolutely. The price/performance ratio is what I'm really interested
in. We, with our 40 server order are not out to impress any
benchmarks, neither are we going to saturate our eth backbones if we
err in the direction of more nodes. Also, our hardware retirement
cycles are pretty slow (I still have some 4 year old machines here) so
I expect these nodes to keep crunching for a while ahead.  In the past
my experience is that users are usually happier with more, slightly
slower nodes than fewer blazing fast nodes. I guess, that's because of
the nature of our codes in that they are very easily paralliziable.

Again to me the few most important parameters seem to be
price/performance and general reliability. A CPU, memory or
motherboard with "problems" (does that still happen these days or is
CPU reliability a given?) is worse than a slightly slower but more
reliable CPU.

-- 
Rahul



More information about the Beowulf mailing list