Anonymous
Not applicable

Why not just run all the jobs on the same cluster? That will save you a lot of time not starting up 20 clusters.