Performance analysis of parallel machines using multi‐read memory

Abstract

AbstractThe multi‐read memory is a special type of shared memory which provides multiple data copies and allows multiple PUs to read data simultaneously. This paper analyzes the behavior of parallel machines using the multi‐read memory and provides some methods to determine their performance. First, a Markov chain model is applied to n ‐ n systems in which the number of copies is the same as that of PUs. Then a less complex expectation model is introduced as well as approximate expressions for the performance and it is shown that the results are almost the same as those of the Markov chain model. The paper also analyzes n ‐ m systems in which the number of copies is less than that of PUs and obtains approximate expressions for the performance. The results of the behavior analysis clarify the upper limit of the number of PUs, the optimum number of copies and other useful results for implementation of the systems.