>> In reference to file staging Slurm has sbcast >> (http://www.schedmd.com/slurmdocs/sbcast.html) that can be used to >> broadcast your files out to your home directories on your compute nodes. > > But not in the simple way as Torque's -W stagein/stageout option. > It is usable mostly for distributing files from a computing node to > others in the same parallel job. > If I want to distribute files from submission node to computing nodes it > requires usage of salloc or shared file system as staging area. > The same is also for sbatch output, which stays on computing node, where > the job was executed or should be redirected to shared area.
Ok, so overall slurm requires shared homes. There is no easy way to do the non-shared home part. I'm contemplating trying to move jobs to use /scratch/... instead of /home/... and then create a smaller /home that is for local users who can then also compute using the scratch space or straight on /home. If we can set this up, then we'll eliminate the need for stagein/out, but right now a lot of scripts rely on it and therefore cannot easily be converted to sbcast usage. We'll see how it goes. In any case it won't be a simple swap exercise to swap torque for slurm hoping the commands work :) Mario Kadastik, PhD Researcher --- "Physics is like sex, sure it may have practical reasons, but that's not why we do it" -- Richard P. Feynman