Ac 2007-2445: High-performance Computing Student Projects

Commodity High Performance Computing (HPC) platforms such as Beowulf Clusters provide excellent opportunities to engage students with challenging projects. Courses such as parallel programming, distributed systems, operating systems, and networking can benefit from the lowcost HPC platform. In this paper we report the results on series of student projects in an advanced operating systems course which jointly have contributed to a larger group project. Several students designed, implemented, and tested segments of manageable term projects contributing to the student learning in the advance topic of high performance computing. We focused on job scheduling for a cluster of processors as the main topic, while pursuing other HPC-related areas such as parallel programming, load balancing, computer simulation, and performance analysis embedded in the theme. In this paper we examine following scheduling polices: FCFS (First-Come-First-Serve), Backfilling Algorithms (Aggressive, Conservative, Multiple Queue, Look-ahead), Coscheduling, and Gang Scheduling. While most of the scheduling policies are batch, Gang Scheduling provides a timesharing approach to the multiprocessor system. Our results indicate that Gang scheduling offers an attractive solution to the drawbacks of batch scheduling. This is especially true with respect to the response time and overestimation of the processing time of the submitted jobs in the system.