Performance evaluation of a cache-coherent multi-processor by iterative mean value analysis

An iterative mean value analysis (MVA) algorithm is developed in this paper for a class of queueing networks (QNs) with forking (MVAF). The algorithmic development is motivated by the performance evaluation of a Multicube multiprocessor system and its cache coherent protocol proposed by Goodman and Woest, 1988. We model the system as a multichain, multiclass, mixed QN with forking phenomena. The MVAF algorithm that we propose approximates a forking by independent flows at each iteration and iteratively applies MVA to the approximate QNs without forking. We prove that MVAF converges to the exact performance measures for the QN with forking. Comparisons among performance evaluations by MVAF, simulation and the mean value performance analysis by Leutennegger and Vernon, 1988, demonstrate both the validity and effectiveness of MVAF.