Study Queuing Networks Using CART.

Y Nianmin,JW Shu,WT Sun
2005-01-01
Abstract:Complex queuing networks such as the fork-join structure usually have not the product result. To get their performance evaluation, we often use simulation method or to build an analytical model which can approximate the performance result. These two method all have their own shortcomings. For example, simulation method is resource intensive and it may take a long time to get a result. And the analytical methods need much expertise and their results usually have perceptible error. In this paper, we try to study the complex queuing networks by CART which is a machine learning tool. After being trained by the performance data gotten from simulation method, the fitted binary tree can be used to predict the performance of the queuing model when inputted by workload parameters. We do some experiments on M/M/1 model and the fork-join structure. The results show that this method can have little average relative error while dealing with them. Using this method needs not much expertise and can predict the queuing network's performance very quickly. It may be a promising method to study the difficult queuing networks.
What problem does this paper attempt to address?