Using Julia with HTCondor

There are two ways to use Julia parallel processing on C4.

1. Run a few workers on a local machine (head node or compute node). The command julia -p n provides n workers on the local machine. If you are just playing around, I would recommend this.
2. Run Julia interactively using the HTCondor ClusterManager interface.

We have extended the Julia ClusterManager interface to support HTCondor on the C4 cluster, so that you can interactively use Julia on the head node and still run worker processes on the compute nodes. To do this, you will need the ClusterManager package; get it by typing

Pkg.add("ClusterManagers")

Once you have installed the package, you can add workers as follows

using ClusterManagers
addprocs_htc(np)  # Where np is the number of processes

The scripts and submission files generated by the addprocs_htc file will be stored in a hidden subdirectory of your home directory called .julia-htc. If there are problems with starting up the workers, you should look there (and in particular look at the standard error outputs from the workers, which are stored in the files labeled julia-XXXX-X.e).

In principle, it should also be possible to do something non-interactive, possibly using the -machinefile argument to Julia inside a script similar to openmpiscript. I have not tried this, though.