Re: [lug-nuts] Re: Supercomputing stuff at Lawrence Livermore Lab

From: Scott Tyson (tysons@deepwell.com)
Date: Mon Dec 20 1999 - 12:14:32 PST


Is this a project to augment the Beowulf Linux project?

----- Original Message -----
From: "William Saphir" <wcsaphir@lbl.gov>
To: "Brian Lavender" <brian@brie.com>
Cc: "Lug Nuts" <lug-nuts@saclug.org>
Sent: Monday, December 20, 1999 11:27 AM
Subject: [lug-nuts] Re: Supercomputing stuff at Lawrence Livermore Lab

>
> Brian,
> Thanks for the pointer to the article.
>
> We are indeed looking for student interns (summer and otherwise)
> to hack Linux. We also have a couple of open staff positions
> (http://www.lbl.gov/CJO/NE11372.html and
> http://www.lbl.gov/CJO/NE11502.html).
> And just to clarify, I am actually from Lawrence *Berkeley* Laboratory
> (we are often confused with Lawrence Livermore) where we have a
supercomputer
> facility (www.nersc.gov) and happen to have the same type of IBM
supercomputer
> discussed in the article, as well as a Cray T3E with 640 Alpha
processors.
> Berkeley's IBM SP is currently at about 300 2-processor machines, and
> will be going to about 150 16-processor machines next year. The
difference
> between Livermore and Berkeley, other than location, is that the
> main focus for Livermore is defense (nuclear bombs) while the main
> focus for Berkeley is basic scientific research (nine Nobel prize
> winners have worked at Berkeley Lab).
>
>
> The Linux clusters we're building are seen as a
> possible long-term replacement for the IBM-style (proprietary)
> supercomputers. The architectures are similar -- both are clusters
> of Unix workstations connected by a switched network -- but the
> IBM machine runs AIX (yuck!) and has a proprietary network.
> Machines assembled from commodity parts and running Linux will
> hopefully be much less expensive and more functional (though
> Linux cluster software is still behind, which is why we have
> a project to develop cluster software for Linux).
>
> Bill
>
>
>
> Brian Lavender wrote:
> >
> > Check B2 of today's (December 20, 1999) SF Chronicle for an article
> > about supercomputing application development at Lawrence Livermore
> > Labratory. I went to a talk a couple weeks ago given by Bill Saphir
> > wcs@nersc.gov at Sac State regarding the use of Linux in Super
computing
> > applications. Interesting thing is that you can effectively build a
> > small cluster up to about fifteen machines using fast ethernet.
Tulip
> > based ethernet cards provide some of the best performance. The
addition
> > of Gigabit ethernet (21, Saphir) provides negligble improvement in
> > performance. But the use of Myrinet showed substantial improvement
for
> > clusters exceeding fifteen machines. These conclusions were drawn
for
> > clustering applications that require heavy interprocess
communication.
> > One such application would be performing a Fast Fourier Transform.
> >
> > The talk was quite interesting, and the impression I got is that
Linux has
> > a predominant future at Lawrence Livermore Labs. In fact I believe
they
> > are seeking interns. According to the words I heard from Bill
Saphir,
> > "The ideal candidate would likely have home based "Linux Hacking"
> > experience." In other words, one who has set up and configured a
linux
> > system at home. And of course, I believe you would have to be
enrolled
> > in some sort of college curriculum.
> >
> > brian
> >
> > Biblio:
> >
> > Saphir, Bill, Linux for Scientific Computing, December 9, 1999
> > http://www.nersc.gov/~wcs/recent_talks/linux4scientificcomputing.pdf
> >
> > New Supercomputer Packs Powerful Punch, December 20, 1999.
> >
http://www.sfgate.com/cgi-bin/article.cgi?file=/chronicle/archive/1999/1
2/20/BU62963.DTL
> >
> > --
> > Brian Lavender
> > http://www.brie.com/brian/
>
************************************************************************
****
> * To UNSUBSCRIBE from the list, send a message with "unsubscribe
lug-nuts"
> * in the message body to majordomo@saclug.org. Please direct other
> * questions, comments, or problems to lug-nuts-owner@saclug.org.
>

****************************************************************************
* To UNSUBSCRIBE from the list, send a message with "unsubscribe lug-nuts"
* in the message body to majordomo@saclug.org. Please direct other
* questions, comments, or problems to lug-nuts-owner@saclug.org.



This archive was generated by hypermail 2b29 : Fri Feb 25 2000 - 14:29:09 PST