Mazouzi,

Watch out, though.  If setting h_vmem, make sure to set a value for
h_stack, as well.  Many codes will fail to start if h_stack is not set
high enough.  I found that setting it to 128m works for Matlab.  I had
to set it to 256m for a cluster that runs a lot of Python.

Setting a default for h_vmem may anger your users, because it is one
more thing they have to check and set.  You can spin it as providing
optimal memory utilization and stable nodes that do not swap/thrash/crash.

Cheers.

On 04/19/2011 05:34 AM, Mazouzi wrote:
> Respect Reuti :))
> 
> On Tue, Apr 19, 2011 at 11:31 AM, Reuti <[email protected]
> <mailto:[email protected]>> wrote:
> 
>     Hi,
> 
>     Am 19.04.2011 um 11:09 schrieb Mazouzi:
> 
>     > Wich solution are you using to share memory between jobs in a
>     single node ?
> 
>     AFAICS these are two different things. You can use implement it by
>     threads or shmem between forks.
> 
> 
>     > We need to avoid using swap.
> 
>     
> http://gridengine.info/2009/12/01/adding-memory-requirement-awareness-to-the-scheduler
> 
>     -- Reuti
> 
>     > Regards,
>     > _______________________________________________
>     > users mailing list
>     > [email protected] <mailto:[email protected]>
>     > https://gridengine.org/mailman/listinfo/users
> 
> 
> 
> 
> _______________________________________________
> users mailing list
> [email protected]
> https://gridengine.org/mailman/listinfo/users

-- 
Gavin W. Burris
Senior Systems Programmer
Information Security and Unix Systems
School of Arts and Sciences
University of Pennsylvania
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to