The effect of average parallelism and CPU-I/O overlap on application speedup

Parallelization of I/O using multiple disk drives is warranted for achieving high performance on systems running CPU and I/O intensive applications. This paper is concerned with the characterization of applications with parallel I/O and the derivation of analytic bounds on speedup. New characteristics for capturing parallelism in I/O are introduced. These characteristics are used in conjunction with other characteristics available in the literature for the derivation of the bounds. Both parallelism in CPU and I/O operations in applications as well as the degree of overlap between CPU and I/O are observed to have a strong impact on speedup.