It is important for almost all transaction processing and computer-communication systems to satisfy response time quantile targets. This paper describes HYDRA, a scalable parallel tool for the analytical determination of response time densities in large, structurally-unrestricted Markov models derived from high-level specifications. The tool exploits an efficient distributed uniformization-based algorithm, combined with hypergraph partitioning to balance computational load across processors while minimising communication. We demonstrate our tool on a 1.6 million state Generalized Stochastic Petri Net model of a flexible manufacturing system, comparing the accuracy of our results with simulation and contrasting the run-time performance of our technique with an approach based on numerical Laplace transform inversion.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.