PBS/Scyld. Re: queue system on Scyld (Thomas Clausen)
newt at scyld.com
Tue Feb 20 13:37:40 PST 2001
> Once this problem is resolved, will this be part of pbs or part of
> Scyld? Will you keep up with the pbs development and always provide latest
> Personally I dont think that is the right approach to modify all existing
> beowulf software available to work with one single system. If the software
> is there, that's fine but what about all the other good stuff out there?
> What if I use LSF? Or Codine which is freely available from sun and should
> be released as open source soon? What about DQS, NQS etc. etc.?
Basically, most software that knows how to manage large SMP or T3E like
systems has no problem on our system.
Surely we would not fault the Cray T3E for providing a comprehensive
system status library and job spawning facilities that don't map well onto
the 1 mom per node model. I feel the same way about our system.
(In fact, we started with the T3E port of OpenPBS but found that it was
not really actively maintained).
The systems most likely to have problems with the Scyld software are
middleware packages that are cluster-centric and depend on having a
poorly-coupled MPP and that further try to address those problems
with solutions. Scyld's product makes many of these problems just go away
in the first place.
Given all of this --- our work to make PBS go in the short term consists
mainly of a generic stub library that can be preloaded or linked against
other systems to help those work as well. We will be producing application
notes for other middleware platforms as our (time) resources allow. We
also look forward to seeing application notes from users exchanged and
discussed in this forum.
Scyld Computing Corporation
More information about the Beowulf