[rescue] Linux help (was ZX coolness)

[Patrick Giagnocavo +1.717.201.3366]

On Tue, Oct 08, 2002 at 09:38:41PM -0400, Al Potter wrote:
> Dave:  as an afficianado of high-end computing, I would have though you got 
> this.  They class of problem they typically work at LANL (Los Alamos 
> National Labs, ie Nuke Central) is typically classed as "embarrasingly 
> parallel" and lends itself very well to the type of clustering possible 
> here.  It really is a big win for them, and IIRC, they built their first 
> Beowulf cluster (around 80 2x proc Alphas) in 1997 or 1998.  It's definitely 
> NOT a fad or marketing-oriented decision.

According to http://loki-www.lanl.gov/ their cluster was late 96/early
97.

And, it was 16 PPro's which delivered 1.2 Gflops, a later version with
32 delivered 2.19 Gflops.

About a year later (June 1998), they did Avalon :
http://cnls.lanl.gov/avalon/ which has/had 140 Alpha CPUs which cost
$313K.  Depending on the software running, they get between 17.6
(treecode) and 47.7 Gflops (Linpack benchmark).

Apparently the earlier versions caught on because Dave Becker improved
the ethernet code in Linux to do channel bonding, which helped a lot
when they were running at only 10Mbits/s .

Cordially
-- 
+--------------------------------------------------+
| Patrick Giagnocavo, patrick at zill.net             |
| Zill.Net - OpenACS, Postgres, Python hosting     |
| OpenACS v4 shared server       $39.99/month      |
| Colocation w/30GB transfer     $99.00/month      |
| Managed servers (incl. system) $175/month        |
+--------------------------------------------------+