Hello again,

Thank's for your responses Daniel and Markus.

For Mr Daniel, I agree with you submitting arrays might be a solution, I
will test it and I'll keep you updated.

For Mr Markus, I think this solution is possible, but I am afraid if I
consider it, I will no longer have a system that ensures a continuity of
service (I mean, if the hardware containing the slurm.conf that controls
the two clusters goes down, the whole system goes down), what's your
opinion on that Mr Markus?

I identified a function that sets the job id (in *job_mgr.c*), this
function is called *_set_job_id* .Now, I will have to figure out when this
function is called by *sbatch.c* or *salloc.c* so as we can see if it's
possible to set up a unique job_id in the two clusters.

I also read an interesting article by *Stephen Trofinoff* entitled « *Exploring
the Implementation of several Key Slurm Inter-Cluster Features*», and in
this article it explains that there's an option (*--sicp*) that handles a
job id from a special reserved range, which means that the job will receive
a unique job id common between the two clusters, It doesn't exist in slurm
14.11.7 does it?


 best regards,
Biki

Reply via email to