On Fri, 10 Jan 2014 17:42:37 +0100
Reuti Reuti wrote:

> Hi,
Hi Reuti,
 
[...]
> Correct. There is nothing in SGE to control this.
> 
> 
> There was the project Gold:
> 
> https://www.mail-archive.com/[email protected]/msg06346.html
> 
> and William integrated it into SGE IIRC. I on my own saw:
> 
> http://gridengine.org/pipermail/users/2011-February/000141.html

Gold!! I took a look at it when I used Torque/MAUI but I don't remember
why we refused to use it ... I'll take a look and ask in mailing list
(still active!!).

> and never investigated it further (as for your own solution:
> prohibiting the submission might not be the proper action (personal
> opinion or course) as the already submitted jobs might need less CPU
> time than requested and others could be run too).
> 
> BTW: who will pay for e.g. reserved but not used cores for a parallel
> job (in case no small job for backfilling is available) could be
> another question.

Yes, there are many situations that we must take in mind... but as a
first approach I chose the simplest solution: if the project used more
than X time, jobs submitted requesting that project should be rejected,
and queued jobs must be change to hold status. 

> NB: Manual for Gold can now be found here:
> http://docs.adaptivecomputing.com/ =>         Archived Documentation
> 
> -- Reuti
> 
> PS: RFE could be a job-start-verifier to check constraints at a later
> point in time
Sorry, Reuti, what is RFE?

Many thanks for your help!
Cheers,
Arnau
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to