Reliability analysis of parallel processing systems
Richard H. R. Harper · 1988
Parallel processors promise an increase in reliability as well as throughput. A quantitative framework for the evaluation and comparison of reliability-enhancement techniques for parallel systems must be employed by system architects. This paper presents a reliability evaluation of a parallel processing ensemble computer as a function of several fault tolerance techniques which might be applied to such a computer. For each fault tolerance approach, Markov models and combinatorial models are used to compute the reliability and expected value of the number of processing elements as a function of mission time. Using this analytical framework for the development of a high-reliability and high-throughput architecture, reliability-critical architectural parameters are identified, leading to a new organizational approach for obtaining high ensemble reliability. The reliability of an ensemble constructed using this approach is computed and found to be superior to the other approaches investigated.