Response time analysis of repairable multiprocessor systems

In this paper, a simple model to analyze a repairable multiprocessor system using queuing theory is introduced. A closed form solution for the average response time is derived and asserted through simulation. The analysis is done for both homogeneous and heterogeneous multiprocessor systems. Unlike homogeneous systems, processors and/or links speed in heterogeneous systems are not the same. We study the effect of processors speed and links speed on the average response time. We also show how the topology of interconnect ion network may affect the average response time. Two mechanisms to repair faulty: processors are considered. The immediate repair mechanism gives bet ter average response time than the delayed repair mechanism. An arriving ta sk to a repairable multiprocessor system may not find the number of availab le processors that give the maximum parallelism and so a minimum response t ime. Waiting until that number becomes available (repaired) will increase t he waiting time. An algorithm is introduced which gives the optimum number of processors that a task must wait before being executed. Waiting for this number will maximize parallelism without excessive increment in waiting ti me.