Response-time simulation of multivariate point process models for multiprogrammed jobstreams

Abstract

We consider simulation methods for particular marked multivariate point processes, called R-processes, which have recently been proposed as models for multiprogrammed jobstreams. Using "workload marks" on events, such models facilitate the incorporation of realistic workload characteristics into computer system performance predictions. We consider R-processes in which workload marks for an individual jobstream form a stationary sequence of discrete random variables having a (generally non-Markovian) mixed moving average-autoregressive dependency structure. For such models we provide a method for obtaining from a single simulation run point and interval estimates for general characteristics of job response times. © 1980 Plenum Publishing Corporation.