Risk of extreme events in the configuration of priority systems

This paper presents a reliability-based framework for the configuration of priority systems that control the risks of extreme transaction latencies. Modeling of priority systems can be computationally complex and lack robustness with respect to the feasible variations of such systems. The analyses tend to be ad hoc and overly particular to the application at hand. Furthermore, modeling of priority systems has focused on assessment of the average latencies, rather than on the potential for extremes of latency. The developed framework includes (i) a compact notation useful for modeling of priority systems, (ii) assessment of the risk of extreme latencies, and (iii) multiobjective analysis for the tuning of system parameters. An acceptable balance is sought among multiple risks of extreme transaction latencies. The framework is demonstrated in the configuration of an e-business server to minimize the lost transactions, e.g. ‘purchase’ requests for which a customer needs to wait longer than 8 s.