[TriLUG] Building a beowulf (or other) cluster

Sean Korb spkorb at gmail.com
Mon Mar 28 22:18:42 EDT 2011


On Mon, Mar 28, 2011 at 10:54 AM, Ron Kelley <rkelleyrtp at gmail.com> wrote:

> Does anyone have a background in building a beowulf cluster or have knowledge in an open-source process management tool we can use?

>From an academic perspective, I really liked working with OpenMosix
http://openmosix.sourceforge.net/

It's defunct, and with good reason: job schedulers turned out to be
more efficient.  But the idea was fantastic and it included a cache
coherent (though slow) file system.  OpenMosix was a kernel addition
that allowed the system to migrate any forked process to an available
machine keeping the grid in perpetual balance.  It wasn't really a
queuing system though and without one it could get hairy if you
weren't sure the number of forks was <= the number of cores available.

It was really neat!

I use straight Sun Grid Engine (kind of like LSF) and it's part of
Rocks.  A little terse to set up if you want to get fancy, but easy to
use in its default configuration.  Every core gets a slot, and then
you qsub a script to those slots (more like "for i in jobs qsub
script.sh").  Jobs are queued until all of them are done.

Hard to tell what Oracle will do with SGE though.  I'm keeping my
fingers crossed.

-- 
Sean Korb spkorb at spkorb.org http://www.spkorb.org
'65,'68 Mustangs,'68 Cougar,'78 R100/7,'60 Metro,'59 A35,'71 Pantera #1382
"The more you drive, the less intelligent you get" --Miller
"Computers are useless.  They can only give you answers." -P. Picasso



More information about the TriLUG mailing list