Fair enough! For what it’s worth, parallel provides a lot of really nice control mechanisms to fine tune how your jobs are scheduled (e.g., only start a new job when there’s X amount of memory available), saving stdout and stderr to log files, running jobs on remote hosts, even saving results to a SQL database.

Well I’m a researcher, so I’m commonly running experiments on lots of inputs. I make scripts to run the experiments that take command line parameters, and then use parallel to run all of my experiments on all of my inputs under all configurations. It’s very nice when you need to try all combos of a bunch of parameters, since by default it’ll run with every combination of parameters you give it.

screen and parallel are my two workhorses.