LHC@home

Greetings!

I am trying to use Boinc in a cluster holding the SLURM batch system. In this way, the jobs interchanged between the Boinc server and the client must be sended through the SLURM batch system to register the entry, and can not be interchanged between Boinc server and client.

Does anyone know how to do this?

Hi,

That's an interesting idea, but I'm not sure how it would work. You could do some tricks to download the WU then submit them as batch jobs to SLURM, but then you are kind of re-inventing a computing element, so you could just set up a computing element and run like a regular grid site instead of using BOINC.

The people here who run BOINC on grid worker nodes run it in parallel to the batch system because it allows over-committing the resources and maximising the overall efficiency of the site. If you submit the BOINC tasks to SLURM it will assign the whole resource to BOINC so you lose this backfilling potential. What is the reason you can't run like this with BOINC in the background of the batch system jobs?

The administrators of the cluster where we want implement the Boinc are very interested in have control on the statistics of jobs, and so, they have imposed the condition of send all jobs through the Slurm system.

We don't really know if this is possible or not, so we are doing a little research about. We will thank any idea coming from the community.

Our second option is, as you said, try to execute the jobs in the batch system with no Boinc.