Hi,

Am 20.09.2011 um 14:30 schrieb Txema Heredia Genestar:

> I have some very important jobs running in my cluster (SGE 6.1u4) that are 
> taking too much time. I wanted to check why they were taking so long and I 
> have found out they where running in the node with a nice value of 19 (the 
> lowest). I have tried to renice them but nothing happened. The most I have 
> seen is when I reniced them while using "top", then I saw the priority 
> modified ( for instance, to -10) for a second, and the next second the 
> priority went back to 19. The only thing I have been able to renice is the 
> sge_shepherd process.
> 
> Is SGE preventing my renice in some way? Is it a known issue?

AFAIR, in 6.1 it was default to have the reprioritization enabled to match the 
functional policy. You can set:

$ qconf -ssconf
...
reprioritize_interval             0:0:0

and

$ qconf -sconf
...
reprioritize                 0

to switch it off.

Nevertheless: only values in the range 0..19 are for user processes. Negative 
ones should be reserved for system tasks. Also note, that the priority in Linux 
is relative. I.e. having 4 processes on a machine (with one core) with all 
having 19 is the same like all having 0. And unless you have more jobs than 
cores on a machine, all will run at their maximum speed anway. In case you have 
a high oversubscription on the machine, maybe you can check the slot count for 
them, or whether there are any jobs running outside of SGE.

-- Reuti
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to