[ale] Linux on a supercomputer

[Vernard Martin]

> The article appears in the July 28, 1997, issue of Government
> Computer News.  Their web site is http://www.gcn.com/.
> 
> Note, they are the ones calling JazzNet a "supercomputer."
> I merely quoted them.

I read the article. They use the age old definiton of supercomputers. "If
it runs as fast as a CRAY, its a supercomputer". The best definition for
supercomputer that I've ever used is "a computer or collection of computers
with highly specialized hardware that is used primarily to achieve very
large scale floating point operations." Most clusters of workstations count 
as supercomputers due to the specialized Networking hardware they use to
"glue" the computers together when acting as one entity. With the fall of
Cray Industries and the death of Seymor Cray, the supercomputing industry
has backed off a bit. The rate of speed increases depends almost entirely
on the middle road of off-the-shelf commoditty parts and almost no new
technology has been developed in the supercomputer only. The most
astonishing research is being done by NEC with their SX-64 vector 
processing machines which currently hold the title of fastest supercomputer.

If anyone is interested in trying to build a cluster of linux workstations
with near-supercomputer levels of performance then I suggest checking out
MPI and PVM first as these are the programming APIs that are most supported
for these sorts of things. If those APIs are reasonable for your coding
effors then you can check into the hardware configurations of things like
JazzNet and Beowulf.

hope this helps

V
--
Vernard Martin, College of Computing Systems Research Laboratory
(vernard at cc.gatech.edu) http://www.cc.gatech.edu/~vernard/
College of Computing, Georgia Institute of Technology, Atlanta, GA 30332
"The truth is a three-edged sword: Your side, their side and the real truth."