Production Throughout as a High-performance Computing Meta-task

Simply put, throughput is the amount of work that a computer or system of computers completes over a given interval of time. Increasing throughput is a constant demand placed on IT managers, researchers, and scientists. Increasing throughput might be accomplished by several different methods, but each method requires scheduling more tasks to machines in the system. However, scheduling tasks in an optimal fashion is very difficult, and in general, has been shown to be NP-complete. Two common metrics for measuring system throughput (i.e., system performance) and scheduler performance are system utilization and final task completion time (i.e., MakeSpan). However, both of these metrics may not reflect the true throughput performance of a scheduler, and hence, the true throughput performance of the system. This paper demonstrates that increasing throughput via intelligent scheduling heuristics can be just as, if not more, effective as other methods. We also present a new metric for measuring scheduler performance, and hence system performance: the Throughput Quotient, T. Sample results demonstrating and contrasting the Throughput Quotient with system utilization and MakeSpan are presented.