Minimizing mean response time in heterogeneous multiple computer systems with a central stochastic job dispatcher.
Keqin Li · International Journal of Computers and Applications · 1998
A heterogeneous multiple computer system consists of a set of autonomous computers with different processing power connected by a communication network. The overall performance (e.g., mean job response time) of such a system can be improved by appropriately balancing the workload over all the nodes in the system. This paper studies probabilistic load-balancing strategies that optimize the performance of a heterogeneous multiple computer system with one job arrival stream and a central job dispatcher. In particular, it shows how to find an optimal job-dispatching method that minimizes the weighted mean response time. We either derive closed-form solutions or give algorithms to find numerical solutions, depending on workload characteristics. One interesting finding is that larger variances of node-processing power result in shorter mean job response time. Furthermore, the coefficient of variation of job execution times has far more effect on the mean response time than the variance of node-processing powers.