[GE users] Re: (GE Users) Job preemption simply unpopular, even if it can be done?
jmarconnet at knology.net
Sun May 8 13:27:31 BST 2005
[ The following text is in the "iso-8859-1" character set. ]
[ Your display is set for the "ISO-8859-10" character set. ]
[ Some special characters may be displayed incorrectly. ]
While awaiting that nugget that would finally help me understand how to do
running job preemption (sorry I'm so dense!) I reread the old and the recent
job preemption posts. I realized how few users have participated in this
thread or in similar earlier ones. Of recent participants, one has a plan,
but has not gotten it implemented, much less tested; one said that
preemption works sometimes, but it is undependable in his experience; and
one has shelved the idea till he gets more time to experiment. So there are
no specific job preemption success stories here lately or in the archives
that I'm aware of. Thank goodness for an easily searchable archives!
I'm coming to the realization that perhaps:
(1) most people here have so many nodes available that there would be little
advantage of persuing job preemption, or
(2) they are not willing to explain how it works to their users, so they
don't even try it,
(3) it does not work dependably, so they ditched it.
So basically the folks here today operate SGE in a FIFO mode once jobs start
running - the way it's really designed to work.
If anyone has a job preemption success story and the info I'm missing to
understand how it is actually set up and what triggers the preemption,
please share it. Yes, I'm aware of the How To for monitoring desktop
computers on a network and using them when they are otherwise unused. We
have a few hundred dedicated nodes with no keyboards or mice to monitor.
Time for me to get off this dead horse and to ask management for a LOT more
nodes so we all can have "our nodes". Perhaps, even better, to dedicate one
node per group as the SGE task manager (or whatever it's really called), so
we each have our own "cluster" that nobody else can see, much less mess with
our specific settings. Since one of our two groups tends to run hundreds of
long jobs (hours or days, sometimes weeks!) and the other group runs tens of
thousands of very short jobs (seconds to minutes), we tend to want different
fundamental SGE settings anyway.
Thanks for the kind, brilliant help along the way. Several good things came
out of this thread that I can use anyway. And others in the future
contemplating the running job preemption concept will have something to read
and to build on.
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.11.6 - Release Date: 5/6/2005
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net
More information about the gridengine-users