We will discuss how to effectively run jobs on Hoffman2 cluster. Hoffman2 cluster serves a wide spectrum of user applications and is almost always heavily loaded. Different kinds of jobs impose conflicting conditions in terms of job scheduling, given available computing resources. We will first review the components of Hoffman2 cluster, the shared cluster model, followed by fundamentals of using it’s job scheduler, Univa Grid Engine. Then a number of use cases will be discussed in details. The materials covered in “Using Hoffman2 cluster” is highly recommended, but not strictly required, prior to attending this class.

Specifically, we will discuss:

• Single-CPU jobs
• Multi-CPU, same-node jobs
• Multiple-node (MPI-style) jobs
• Job arrays
• Techniques to inspect and fix stuck jobs