[GE users] qstat -pri doesn't show priority (scheduler problem)

Sean Dilda agrajag at dragaera.net
Fri Jan 21 18:58:53 GMT 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. ]

Christian Bolliger wrote:
> Hi
> We have a scheduler problem in SGE 6.0 (6.0u3, already in 6.0u2).
> The job priority of the jobs in the pending state are not shown.  But 
> changing the priority of a job manually or putting a job on a hold gives 
> the job a priority != 0.0000.

Running cvs code between 6.0u2 and 6.0u3, and I don't see this. 
However, it may be that the job priority is set at the next scheduler 
run, and I have mine configured to run almost immediately after a job is 

> Furthermore new parallel jobs just overtake jobs with a calculated 
> priority (see above).

Run 'qconf -sc' and look at the 'slots' line.  The last number is an 
urgency value.  By default its something like '100'.  That means that 
for each slot requested the urgency value (before normalization) is 
increased by 100.  Therefore, new big parallel jobs have more priority 
(based on their urgency value) than non parallel jobs.

> Schedule intervall is set to 00:05:00 (256 nodes). Is this a 
> configuration problem or a bug?

Wow, 5 min is a long interval.  I have 274 nodes and my schedule 
interval is set to 00:00:15.  and flush_submit_sec and flush_finish_sec 
are both set to 1.  I also have a dual proc machine dedicated to running 
sge_qmaster and sge_schedd.

 From what I said above, it sounds like you're dealing with 
configuration issues and not a bug.  Try making your schedule interval 
smaller, and set the slots urgency to '0', and see if that gets rid of 
the problems you're seeing.

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 mailing list