about fast interconnects and SCI in particular

[Don Morton]

My "personal" feeling is that you'd be better off with more CPU's
and sacrifice on the communication, particularly, since you seem
to suggest more of a development environment with the "big"
problems running elsewhere.  My own experience suggests that
with several users, you need the CPU's.  One can also argue
that using Fast Ethernet makes the communication more noticeable,
hopefully encouraging users to deal with the issue rather than
ignoring it.

You might also consider that nodes will go down, and if you only
have 16 to start with, you don't have much to play with.

I'm sure others may disagree with me, but I've always felt that
when dealing with Beowulf primarily for development and training
activities, that performance isn't a "huge" consideration.  Like
you suggest, you can always go to the supercomputer centers with
the big problems.  In fact, you have a much better shot at getting
computer time if you already have a parallel code that runs
efficiently.


On Mon, 14 Jun 1999, Florent Calvayrac wrote:

> 
> I have  been involved since October 1998 in the definition, fund raising
> and purchase  of a cluster for computational physics purposes,
> and  we are about to take a final decision on the nature of the cluster
> (processors and communications hardware).
> 
> I had already asked the following question on comp.parallel last year,
> and got various and interesting answers, but am still in trouble :
> 

   Don Morton                  http://www-scinet.umt.edu/~morton/
   Department of Computer Science       The University of Montana
   Missoula, MT 59812 | Voice (406) 243-4975 | Fax (406) 243-5139

   Summer: Arctic Region Supercomputing Center
           University of Alaska