[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Job limit for a condor user
- Date: Wed, 29 Mar 2006 11:59:53 -0600
- From: "David A. Kotz" <dkotz@xxxxxxxxxxxxx>
- Subject: Re: [Condor-users] Job limit for a condor user
I was waiting to see if you got a more insightful response than what I
can provide, but lacking that, here goes...
That user's userprio matters only in comparison to other users competing
for the same resources. If he submits all jobs from one account,
whether from a single schedd or from multiples, and if no other users
have submitted jobs, userprio should be irrelevant. If he submits most
of his jobs from one account, then submits some more from another
account, the second account would have a better priority and might cause
the preemption of some of the jobs from the first account.
I have no idea why some of the machines are not being allocated, but I
doubt it's because of userprio. I don't think I've had a single user
allocate more than 200 or 300 machines, so I can't say that I've seen it
successfully scale to the numbers you're talking about, but I'm not
aware of any reason it should not. You might try using "condor_q
-better-analyze [job number]" to see if it can give you a good reason
that some jobs remain idle. It usually segfaults for me, but it might
work for you. It seems to do the more detailed analyzis that used to be
in the condor_analyze command, at least in the few cases wherein it
Baker D.J. wrote:
One of the users of the condor pool here is submitting jobs from two
different machines. He hopes to be able to increase his productivity by
submitting jobs from multiple machines, however I'm wondering if this is
a good idea and/or whether it actually works. As more jobs from this
user are submitted to the pool does his priority drop to a point at
which he can (for a while) not match any more jobs? I was viewing the
Matchlog on the master at the time, and noted that his jobs were being
rejected with an "insufficient priority" message. Can anyone please help
us to understand the situation here. At the time he managed to match in
the region of 400 jobs to machines, however there were still in excess
of 800 machines still available to do work.
If this user really wants to do the above then should he really use a
different username on his multiple submit machines -- so that he does
not compete against himself for resources?
Thank you -- David Baker.
Condor-users mailing list