Scyld boot, big node_up, run rpm PBS?

[hanzl at noel.feld.cvut.cz]

Did anybody try to augment node_up so as the nodes can run ordinary
OpenPBS?


(Or - would you suggest other batch spooling to be used on Scyld
system now?)


I am just converting our MAGI cluster (based on replicated disks with
RedHat, NFS-mounting just /usr on some nodes) to Scyld bproc-based
system. Scyld way is great, I love it - make nodes beobootable and
never touch them again, no more version skews. (I never used NFS root
being afraid of performance.)

I would also like to use new batch spooling (so far using my own
system which was never finished). PBS seemes good (though I am afraid
that their licensing strategy lessens community support PBS could
receive should it be easier to download).

Now, how shall I install PBS on Scyld system?

I know Daniel Ridge is working on it, I know Greg Lindahl says it
works on CPlant without 1 mom per node - but none of this seemes to be
an easy way for me to get it now.

Would it be possible to screw up node_up to get close to NFS-root
system completness (I know it violates basic idea behind bproc, I
know...) and run an ordinary rpm-installed OpenPBS? (It would also
satisfy my users who are still used to rsh around the cluster etc...)

My batch spooling requirements are actually quite small, I need just
afterok:jobid feature of qsub (job dependency) and till now I somehow
managed with two pages long daemon shell script running my jobs -
maybe PBS is not the only choice?


Thanks for any tips

Vaclav Hanzl