Server-side scheduling in cluster parallel I/O systems

Parallel I/O has become a necessity in the face of performance improvements in other areas of computing systems. Studies have shown that peak performance is infrequently realized, and work in parallel I/O optimization strives to achieve peak performance for applications. In this paper we revisit one area of performance optimization in parallel I/O, that of server-side scheduling of service. With the wide variety of systems and workloads seen today, multiple server-side scheduling algorithms are necessary to match potential workloads. We show through experimentation that performance gains can be seen in practice through the use of alternative scheduling algorithms, but that no single algorithm provides the best performance across the board. Finally we discuss the potential for automatic matching of server-side scheduling algorithms to workloads in real-time