Am 21.02.2012 um 08:58 schrieb Mazouzi:

> On Tue, Feb 21, 2012 at 12:52 AM, Reuti <[email protected]> wrote:
> Hi,
> 
> Hi Reuti,
>  
> Am 20.02.2012 um 21:32 schrieb Mazouzi:
> 
> > I have a doubt about combining mem_free with h_vmem.
> >
> > I set up h_vmem correctly as described here :
> >
> >
> >
> > I would like to combine h_vmem with mem_free that user can request memory 
> > like this :
> >
> > qsub  -l mem_free=MEM_NEEDED,h_vmem=MEM_MAX
> 
> If you want to combine it in the way: sum of mem_free is physical memory 
> installed, sum of h_vmem is physical memory plus swap over-commitment you can 
> do so and set both up in the same way. Usually only one of them is used, as 
> all jobs will be affected by the swapping if it occurs. Nevertheless it might 
> work for you.
> 
> 
> I would like to prevent job to use swap. So I plans to set up h_vmem as the 
> sum of of phisical memory installed. I know that, in this case, h_vmem is 
> sufficient to do that. But I would like to keep mem_free for now. is it a 
> safe way to combine both ?

But how will you set mem_free then as initial value for the consumable - lower 
than the installed memory? As you know that h_vmem will never be passed, I 
don't see any advantage of using it then. I would suggest to remove it.

-- Reuti


> Thank you.
> 
>  
> > do I need to make mem_free consumable ? or just let the SGE sensor update 
> > this value ?
> 
> If you make it consumable, the load sensor will still be used if its value is 
> lower. This is especially useful as the real memory consumption may show up 
> after some runtime of the job only, the consumable is applied instantly.
> 
> The output in `qhost -F mem_free` will show either hc: loder hl: for it, 
> depending on which value is the constraint.
> 
> -- Reuti
> 
> 
> > complex attributes :
> >
> > h_vmem              h_vmem     MEMORY      <=    YES         YES        1g  
> >       0
> > mem_free            mf             MEMORY        <=    YES         NO       
> >   0        0
> >
> > Thank you.
> > _______________________________________________
> > users mailing list
> > [email protected]
> > https://gridengine.org/mailman/listinfo/users
> 
> 


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

Reply via email to