Reduction of computational complexity of mean value analysis in modeling of multiple class-server stations for optimal performance

Abstract

The evaluation of stochastic systems results in product-form queue networks with multiple customers’ class-server sta tions. As a result of the exponential computational complexity involved in the number of classes, it is impossible to evalu ate this model efficiently using the mean value analysis (MVA) algorithm. A number of approximation methods had been applied to minimize corresponding problems of computational complexity but none had sufficiently solved the problem involving multiple servers. We propose a new approach, the Multiple Server-Station Expected Performance Measure (MSEPM) to reduce the computational complexity involved in the use of the mean value analysis for the performance evaluation of multi-classes, multi-server stations using the first-come, first-served policy. The Arena training and evalua tion mode (student) simulator was used to test the viability of MSEPM. Numerical results were compared with the results generated using exact MVA algorithm for the same experiments. Results indicated an improved performance of MSEPM over the exact MVA algorithm.