Workload characterization of the shared/buy-in computing cluster at boston university

Computing clusters provide a complete environment for computational research, including bio-informatics, machine learning, and image processing. The Shared Computing Cluster (SCC) at Boston University is based on a shared/buy-in architecture that combines shared computers, which are free to be used by all users, and buy-in computers, which are computers purchased by users for semi-exclusive use. Although there exists significant work on characterizing the performance of computing clusters, little is known about shared/buy-in architectures. Using data traces, we statistically analyze the performance of the SCC. Our results show that the average waiting time of a buy-in job is 16.1% shorter than that of a shared job. Furthermore, we identify parameters that have a major impact on the performance experienced by shared and buy-in jobs. These parameters include the type of parallel environment and the run time limit (i.e., the maximum time during which a job can use a resource). Finally, we show that the semi-exclusive paradigm, which allows any SCC user to use idle buy-in resources for a limited time, increases the utilization of buy-in resources by 17.4%, thus significantly improving the performance of the system as a whole.