[Bioclusters] Request for discussions-How to build a biocluster Part 4 (batch systems)

[Ron Chen]

> both take a bit more time to customize for things
> like consumable resources and parallel environments
I think SGE is better, since with OpenPBS, you need
to recompile in order to add a comsumable resource
to the cluster.

I used to play with PBS for a short while and then
switched, mainly due to --

1. there is not much failover features in PBS. If your
master node dies, then your cluster stops running. SGE
has 1 or more shadow masters, which can take over if
the master fails.

2. PBS does not scale (there is not much improvement,
I still receive mail complaining about this)
I used to submit 30K jobs to see "what would happen",
but SGE did not crash. The emails on PBS lists
complain about PBS not being able to handle large
number of jobs (5K)

Also, if one of your execution node dies, PBS server
sometimes hangs, waiting for timeout. This does not
happen to SGE, and SGE is smart enough to reschedule
the job to another machine. (It assumes the job is
lost)

> I believe PBSpro eventually folds improvements into
> OpenPBS, so that might not be seen as that great of
> a benefit.

There are enough things that PBSPro has but is missing
in OpenPBS. (mainly my opinion, but also the reactions
from the mailing-list)

The final thing I like about SGE is that Sun is
willing to improve it, but it is open enough that you
can go for other companies to ask for support. This
is the real open source model that I want.

-Ron

__________________________________________________
Do You Yahoo!?
Yahoo! Health - your guide to health and wellness
http://health.yahoo.com