I have not tried it, but I think you can define a queue per user and the make
this user the owner of the queue. This way they could modify their queues, and 
thus 
the number of processes. However it seems insecure, they can change a lot of 
stuff, 
and you have to control the number of jobs per host in another way (maybe a
consumable).

NiCo

Excerpts from Txema Heredia Genestar's message of mié may 15 17:23:53 +0200 
2013:
> El 15/05/13 14:58, William Hay escribió:
> >
> > On 15 May 2013 13:32, Txema Heredia Genestar <[email protected] 
> > <mailto:[email protected]>> wrote:
> >
> >     Hi all,
> >
> >     I was wondering if there is any way to allow a user to choose how many
> >     jobs they want to have running concurrently in the cluster. I am aware
> >     that I, as an administrator, can specify limits in the slot usage for
> >     each user whith resource quota sets.
> >     What I am asking is a method to allow a user to submit, for instance,
> >     2000 jobs, but having only 50 running simultaneously, and, two days
> >     later, be able to run 400 jobs at once.
> >     Currently I am using a consumable attribute set to the total number of
> >     cores of our cluster (400), so users can request some number (400
> >     / 50 =
> >     8) in order to have their desired simultaneous job, but this leads to
> >     some confusion and applies to all users at once (the consumable
> >     attribute pool is shared among all users).
> >
> >     Is there a fast-and-easy way a user can set his own limit?
> >
> > Per user resource quota on the consumable?   Use  some wrapper 
> > scripts/jsv to do the maths.
> >
> What do you mean by "per-user resource quota on the consumable"?
> 
> Do you mean creating a resource quota for the global consumable 
> attribute so a user can only request a given number? That would only 
> make all users compete for the same, with further restrictions.
> 
> The more I think about it, the more I realize of the problems that what 
> I request imply: The "number of simultaneous jobs" cannot be requested 
> by job, as it depends of the system as a whole, and any workaround I 
> think about, implies granting users managing privileges.
> 
> Maybe a solution would be out-SGE: I could create a script in crontab 
> that reads one file where I store all RQS, and also reads, say, 
> ~/.sge_rqs for each user, granting to each user a slots limit of the 
> minimum of the 2 files...
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to