Hybrid performability analysis of logistic networks

Resources in large logistic networks are occasionally unavailable or malfunctioning. This implies that performability becomes an issue for quantitative analysis of logistic networks. Different time scales between failures and normal operation often justify the decomposition of a performability model into a single availability model that considers failures and recovery of resources and a family of performance models whose individual instances depend on the state of resources. In this paper, we present an approach that simulates a set of performance models independently and in a distributed manner on a network of workstations. We propose to optimize the achievable quality of results for a given total amount of CPU time by minimizing the confidence intervals for performability measures. This is possible by an adaptive assignment of CPU time to simulate those models whose results have the largest impact on the width of confidence intervals.