Reduction of Closed Queueing Networks for Efficient Simulation

This article gives several methods for approximating a closed queueing network with a smaller one. The objective is to reduce the simulation time of the network. We consider Jackson-like networks with Markovian routing and with general service distributions. The basic idea is to first divide the network into two parts—the core nodes of interest and the remaining nodes. We suppose that only metrics at the core nodes are of interest. The remaining nodes are collapsed into a reduced set of nodes, in an effort to approximate the flows into and out of the set of core nodes. The core nodes and their interactions are preserved in the reduced network. We test the network reductions for accuracy and speed. By randomly generating sample networks, we test the reductions on a large variety of test networks, rather than on a few specific cases. The main conclusion is that the reductions work well when the squared coefficients of variation of the service distributions are not all small (that is, the network is not close to being deterministic) and for nodes where the utilization is not too high or too low