networking options

[Patrick Geoffray]

On Tue, 2002-09-24 at 17:22, Peter Kjellström wrote:
> On Tue, 24 Sep 2002, Richard Walsh wrote:
> 
> > 
> > All,
> > 
> > Whoops ... down is not alway bad ... in my prior note
> > I meant gets worse (grows larger quickly).
> 
> Message read loud and clear :-) ...but I do disagree. SCI latency is not
> as dependent on the size of the cluster as one might think considering
> it's ring topology. Here are a few numbers (all application level,
> mpi_send/recv):

I agree, the latency is not at first sight affected by the rings
topology. It has however an effect on latency when the load increases,
when bandwidth starts to quick in. 
If you take you car to cross Los Angeles to go to LAX airport, you can
do that in 20 minutes at 4am. At rush hour, when everybody is roughly
trying to do the same thing, it's about 3 hours. When you share links,
you are exposed to such load effect.

The first post was talking about 1000 nodes, that's a lot of cars.

> Thus, running a 512 node SCI 3d torus (8x8x8 ring dimension) would give

Much more important for this cluster size (1000 nodes) is the physical
cabling and reliability of the whole thing. You cannot use any
interconnect today to build 512 nodes, and pardon me but you will have
to be nuts to do it with a Torus.

My 2 things.

Patrick
-- 
----------------------------------------------------------
|   Patrick Geoffray, Ph.D.      patrick at myri.com 
|   Myricom, Inc.                http://www.myri.com
|   Cell:  865-389-8852          685 Emory Valley Rd (B)
|   Phone: 626-821-5555          Oak Ridge, TN 37830
----------------------------------------------------------