50 GFLOP LinxBIOS cluster up (fwd)
Eugene.Leitl at lrz.uni-muenchen.de
Thu Jun 28 04:29:07 PDT 2001
-- Eugen* Leitl <a href="http://www.lrz.de/~ui22204/">leitl</a>
ICBMTO : N48 10'07'' E011 33'53'' http://www.lrz.de/~ui22204
57F9CFD3: ED90 0433 EB74 E4A9 537F CFF5 86E7 629B 57F9 CFD3
---------- Forwarded message ----------
Date: Wed, 27 Jun 2001 22:11:57 -0600 (MDT)
From: Ronald G Minnich <rminnich at lanl.gov>
To: linuxbios at lanl.gov
Cc: Daryl W Grunau <dwg at lanl.gov>
Subject: 50 GFLOP LinxBIOS cluster up
Daryl Grunau (from here) just finished bringing up 100 DS10L (466 Mhz.)
Alpha nodes. The sequence was to boot them into Linux as NFS-root nodes,
run a script to reprogram flash (from Linux) with linuxbios+scyld, then
reboot. Boom, up comes another Scyld client. Repeat, painfully, 100 times.
What's interesting is that I've been told by a well-known vendor that it
is impossible to reflash FLASH on an Alpha from anything other than the
SRM (BIOS). Ha. I love doing things that are impossible :-)
The pessimistic performance # for this cluster is 50 GFLOPS (100 * 466
Mhz). Since these are dual issue machines, the marketing number is about
100 GFLOPS, much more impressive but much less honest.
Boot from power-on into fully functioning node is 45-50 seconds. Around 30
seconds of this are the Serial ROM (i.e. something we can't fix). 10 or
for Linuxbios startup, another almost 10 for Scyld.
We can't thank Eric Biederman enough for making the DS10 LinuxBIOS
work. But, Eric, thanks!
More information about the Beowulf