32 + 8 nodes beowulf cluster design.

[TIERNEY CRAIG CRISTY]

I would be interested in your problems with Myrinet
and why due to stability reasons you did not choose it.
There are many sites that use Myrinet very successfully
(Including here at FSL).  It would cost much more, and
that is a good reason not to choose it if your codes
scale well with ethernet.  You are going to be limited in 
running jobs on the whole cluster with just one ethernet 
interface for communication.

I think you may be pricing the Portland Group software
incorrectly.  If you just want the compilers, you can
purchase it for just the front end.  Unless someone knows
differently, their MPI is not any more efficient that
mpich or lam.  The parallel tools don't do much good
as you only have single CPU nodes.  Therefore HPF is
overkill.  The compilers are good though, as I have used
them on several small x86 clusters and have been quite
happy with the performance.

If you have extra money, I would consider getting
serial port switches for the nodes.  You can debug
nodes remotely if the ethernet is down.  Having a 
terminal server is very handy.

Craig

Craig Tierney (ctierney at hpti.com)


-----Original Message-----
From: Paul Maragakis
To: beowulf at beowulf.org
Sent: 2/12/01 8:14 PM
Subject: 32 + 8 nodes beowulf cluster design.

Hi everyone,

I have just drafted a proposal listing the hardware components and their

prices for a 40 nodes beowulf using 1.2 GHz Athlons, 8*1.5 GB + 32*512
MB
memory, a double fast-ethernet network, and a front-end for storage and
administration.  The draft is at:
http://hdsc.deas.harvard.edu/~plm/beowulf.pdf
All comments are highly appreciated, especially those regarding the
network, the UPS and the price estimates.  

Take care,

Paul