Bill Saphir from Lawrence Berkeley Labs gave a talk on linux clusters (or clusters in general).
<andy's paraphrase, with apologies to Bill>
We do scientific computing. We don't care what OS. We just want to do science, not computer science.
We use linux as an insurance policy. It's there, we can fix things when they break and we can
create clusters.
</end paraphrase>
Among the most notable quotes:
" 64 bit floating point calculations is everything (well almost) in scientific computing. "
" a Cluster is a group of computers used as a single resource."
There were lot's of questions (most of which I didn't understand) from the group. Which amazed me
in their scope of their knowledge. There was also a fair amount of discussion about network cards.
Bill spoke about how clusters evolved. It boiled down to speed and bottle necks. (If the bus is the
bottle neck then I'll add more machines), mentality prevailed. Now we have access to multi-processor
nodes in a cluster of machines where the bottleneck is the network instead of the bus speed. All
in all it was very informative.
During the break and after the meeting:
There was lot's of discussion about dsl. Poor Mike Machado was innundated with questions about innercites
dsl.
Ther were 30 people at the meeting. It spritzed as we were leaving (I guess it is still spring/winter in Sac.)
Links:
[pdf]* info about clustering by Bill Saphir.
[NERSC]* (National Energy Reasearch Scientific Computing Center) web site.
[BDI]* Berkely Lab Distribution.
[Beowulf HOWTO]*
[CESDIS beowulf]* NASA's cluster/
[beowulf-underground slash like beowulf info site.
[Debian Beowulf]* Just thought I would throw that in.
* offsite links are displayed as [link]*
more next month.
-- Andy
Back to the Saclug.org home page.
Brians announcement:
Bill Saphir will speak about Linux clustering and their efforts at Lawrence Berkeley Labratory.
Linux clustering is where you network multiple machines together with Ethernet, or other networking
protocol such as Myranet, and you structure the machines together so that they can perform tasks and
the processes can communicate amongst each other. Linux clustering has demonstrated super computing
performance on scale with the massive Super Computers such as the Cray. Clustering also demonstrates
certain advantages over having a large monolithic supercomputer. Even if you are not a super computer
scientist, this will be a hot meeting. Bill gave a presentation at Sacramento State in December 1999,
and after listening to his talk, I felt that even I could develop a super computer cluster. Get the
meeting announcement in pdf or postscript format, and tell your friends. This should be a hot meeting.
Previous Meetings.
Mar. 2000
Oct. 1999
Sep. 1999
Jul. 1999
Jun. 1999
May. 1999
Apr. 1999
Mar. 1999
Feb. 1999
Jan. 1999
Dec. 1998
Oct. 1998