NodeFile and qsub script

[juliaparallel]

NodeFile and qsub script

[amitmurthy]

cc @nlhepler

[floswald]

hi,
chasing up on the issue brought up by @tlamadon, here is a repo with a working version for a pbs cluster.

https://github.com/floswald/parallelTest/tree/master/julia/iridis

same applies: on a big machine it's infeasible to run jobs interactively, hence the need to first tell qsub that you want x machines and y processes, and then once the scheduler is finished and you have to a resource allocation in a PE_NODEDILE or similar, you do addprocs(m) within a julia script where m is a vector of machines names you suitably parse from PE_NODEDILE.

It seems there should be a distinction between interactive cluster sessions, and batch ones.

[juliohm]

Too old to reproduce. Closing.