[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] environment = "OMP_NUM_THREADS=$$(CPUS)" don't work with dynamic slot



Hi Frédéric:

I do not see this behavior in the version of Condor that we are running (7.5.4) - OMP_NUM_THREADS=1 whether it matches with a dynamic slot or not. It is possible this behavior was changed in the development series, but has not yet made it into the stable branch.

As an alternative - provided you are always specifying 'RequestCpus' in your submit file, then

RequestCpus = 1
Environment = "OMP_NUM_THREADS=$(RequestCpus)"

should work regardless of whether you match with a dynamic or static slot.

Alternatively, if you don't want to have to specify RequestCpus in your submit file, you should be able to specify 'Environment' like this:

Environment = \
"OMP_NUM_THREADS=$$([ifThenElse(isUndefined(MY.RequestCpus),1,MY.RequestCpus)])"

This will look through the job classad to see if RequestCpus is defined. If it is not, OMP_NUM_THREADS will be set equal to 1 (in the case where the job matched with a "static" slot), otherwise it will be set equal to the value of RequestCpus (i.e. the job matched with a dynamic slot).

Best of luck,
David


On 01/22/2011 12:35 PM, Frédéric Bastien wrote:
Hi,

In my cluster I have 1 node in dynamic partionning mode to test it.
I'm using condor 7.4.2. In the submit file I set the environment
variable OMP_NUM_THREADS to the number of CPU in the slot.

environment    = " OMP_NUM_THREADS=$$(CPUS)"

When a 1 cpu job go to a non partitionable slot, all work correctly. But
when that same jobs go to the partitionable slot, the environment variable
OMP_NUM_THREADS got set to the number of available slot in the dynamic
partitionning node before that jobs start.

Example. the node have 8 cores. So the first 1 cpu jobs that start on
it have OMP_NUM_THREADS=8, the second 1 cpu jobs have
OMP_NUM_THREADS=7, ... It seam the substitution of "$$(CPUS)" get done
on the class
ad of the partitionable slot and not on the created slot!

Do anybody have any hint on how to solve this? Do you see that behavior?

Their is nothing mentioned in the release note of condor 7.4.3 and
7.4.4 that tell they fixed something like that.

thanks for your help and for the great product that condor is.

Frédéric Bastien
_______________________________________________
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/