Hi Dave,
On Fri, Mar 1, 2013 at 7:04 PM, Dave Love <[email protected]> wrote: > Raymond Wan <[email protected]> writes: > >> RQS sounds like what I'm looking for (for #1) -- the problem I'm >> having is that users are being a bit greedy when I'm not watching. > > It doesn't seem to have been said that RQS only affects scheduling the > job, not what happens at run time. It specifically won't prevent a job > using more cores than it's asked for (assuming slots≡cores). Thank you for the two follow-up messages. Do you have a suggestion on what I should do? Basically, I have a problem with users over-subscribing. I'm asking them that they can use only n cores and they are *purposely* using more than that. The number of cores being used added to the number of cores in use by other users at the time is more than the number of cores the system has. I can have jobs suspended when the server's load gets too high, but is there anything else I can do to ensure users are playing fair with each other? As to your other question about queues, currently, I created queues and set the number of processors. i.e., $ qconf -sq q1 | egrep processors processors 1 I thought that setting would do it, but it doesn't seem to do anything. And looking at the man pages make it look like it isn't doing what I think it is... > You really want core binding, but as Reuti says, how a multi-threaded > job uses the assigned cores is up to it, if it even chooses to obey it. > Typically you set something like GOMP_CPU_AFFINITY (for GNU OpenMP) > defined -- see past posts here. Ok! Thanks for giving me something to search for! I'll look into it. My priority is still the first problem since I don't know how to deal with users who aren't sharing... Ray _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
