[GE users] Parallel environment and flexlm

Reuti reuti at staff.uni-marburg.de
Thu Aug 30 12:32:13 BST 2007

Am 30.08.2007 um 13:02 schrieb Colin Thomas:

> We are using 6.0u9.
> We have set up a parallel environment "calibre", with $pe_slots to
> ensure it goes onto just one machine.
> We have a job we want to reserve all 8 cpus on a machine for this job,
> so we run "-pe calibre 8" (there is one slot per cpu in this queue)  
> 9The
> calibre job will multi thread on all cpus on the allocated machine).
> Now we also use the oleson flexlm method, and have tokens  
> calibrelvs and
> calibrehlvs
> For a 8 slot parallel job we run
> -pe calibre 8 -l calibrelvs=1,calibrehlvs=1
> But it goes into pending, with
> "can not run in PE "calibre" because it only offers 0 slots".
> After some debugging, we only have 7 calibrehlvs tokens. If we set  
> "-pe
> calibre 7" it will run.
> This therefore implies that the number of licenses the grid is looking
> for is "number in the -l" * "number of pe slots". (And also the  
> message
> seems to be misleading (perhaps)).
> This is not the case though, I want just the number in the -l  
> definition
> to be consumed and no more.
> I have looked at the PE definition, and see no way to tell the grid  
> not
> to multiply resources up :-(

This is already an RFE:


> I would be grateful of some pointers on where I might be going wrong.
> I look forward to your reply.

For now you can use 0.125 as resource request, which will be  
multiplied by 8.

-- Reuti

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