Re: [Condor-users] Two pieces of software with different threading models using the same pool

[David Kotz <dkotz@xxxxxxxxxxxxx>]

Neil,

If I understand your problem correctly, for your first job type, you
want each job to claim an entire machine exclusively, and for your
second job type, you want to allow up to NUM_CPUs/2 to run on a machine,
as long as there are no jobs of the first type running.

In that case, you can refer to this thread from the list archive:

https://lists.cs.wisc.edu/archive/condor-users/2007-June/msg00295.shtml

You'd want to define two slot types:

Type 1 has all CPUs, memory, etc. on the machine.  NUM_SLOTS_TYPE_1 = 1.
Type 1's START expression evaluates to FALSE if any of the other slot is
claimed.

Type 2 is a mostly normal slot definition, except that each one claims
two CPUs.  NUM_SLOTS_TYPE_2 = NUM_CPUS/2.  Type 2's START expression
evaluates to FALSE if the state of slot Type 1 is "Claimed".


Is that what you're trying to achieve?

- dave



On Thu, 2010-07-01 at 14:30 -0400, Neil Woodhouse wrote:
> Fellow Condor Users,
> 
>  
> 
>                 I have two pieces of software, from different vendors,
> that are Condor enabled. One of which I have some control over. They
> use different threading models one uses all of the CPUs on the node
> and the NUM_CPUs is set to 1, or one slot. The other software, of
> which I can control, knows it’s limitations and uses 2 threads
> optimally. The number of slots is divided by 2 on each machine and the
> NUM_CPUs set to that value. The first piece of software needs to use
> one slot and fails miserably if the number is changed.
> 
>  
> 
>                 Does anyone have an idea of how I may run both
> software programs together and concurrently? I am assuming that I may
> change the requirements. These processes take hours to process, so at
> any one time any slot may be being used. Dividing the pool is a
> possibility, but the downtime for some of the resources may be too
> long and the machines should keep chugging along. 
> 
>  
> 
>  
> 
>  
> 
> Neil
> 
> 
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
> 
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/condor-users/