A Performability Analysis of Two Multi-Processor Systems
Matthew R. Smith, S K Trivedi · 1986
Multiple-processor systems can provide higher performance and higher reliability/availability than single-processor systems. In order to properly assess the effectiveness of multi-processor systems, measures that combine performance and reliability are needed. We describe the behavior of the multi-processor system as a continuous-time Markov chain and associate a reward rate (performance measure) with each state. We evaluate the distribution of performability for analytical models of two multi-processor systems using a recently developed polynomial-time algorithm that obtains the distribution of performability for non-repairable, as well as repairable, systems with heterogeneous components. The systems that we analyze are the (C.mmp) multi-processor system developed at Carnegie Mellon University and a shared-storage system proposed by researchers at the IBM Watson Research Center.