Am 25.07.2011 um 18:54 schrieb Mohamed Adel:

>> any output in the messages file of the qmaster?
> 
> 07/25/2011 19:44:01|worker|batch01|C|!!!!!!!!!! got NULL element for CE_name 
> !!!!!!!!!!
> 
>> There are three things to note:
>> 
>> - Limits specified in an RQS aren't enforced. They only specify what you can 
>> request on the command line with the -l option. Only the specification of 
>> the -l option will enforce the limit then in the end.
> I should have clarified this before. All I want to do is to specify a CPU 
> quota for projects; i.e. SGE projects should have a limited core-hours for 
> the project life time. For example, test project was assigned 6000 
> core-hours, it submitted jobs that consumed 1000 core-hours, then it should 
> have 5000 core-hours remaining.
> From your explanation, it seems that RQS is not the choice to do so.

Correct. You want to implement something like a "budget" for eacj project, 
which they can consume.


> Have you got any idea how can I implement this accounting policy using SGE, 
> or should I turn to writing custom scripts for that?
> Perhaps there are some tools that could be integrated with SGE to implement 
> accounting policies!

Yes, it was called Gold, I don't know whether it's still working and fits your 
needs: http://sourceforge.net/projects/gold/

-- Reuti


>> - The first matching rule will exit the RQS. So most likely the check "limit 
>> users mab to slots=100" succeeds and allows the job to run. The others are 
>> never checked.
> that explain why submitting parallel jobs doesn't crash qmaster.
> 
>> - As a result of this implementation, you will need either two RQS: one for 
>> h_rt and one for h_cpu *or* put them in one rule, separated by a comma.
> 
>> Nevertheless SGE shouldn't crash.
> unfortunately, it did!!
> 
> thanks in advance,
> madel


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

Reply via email to