Using a segregation measure for the workload characterization of multi-class queuing networks
Lawrence W. Dowdy, Alan T. Krantz, Michael R. Leuze · 1990
When a queuing network model of a computer system is constructed, the workload is characterized by several parameters known as the device demands. The demand that every customer places upon every device must be specified. A complete workload characterization of a K device network with N different customers contains N * K parameters. Substantial savings in complexity result if the number of workload parameters is decreased. If, for example, only the average demands on the K devices are used in the workload characterization, the overhead of parameter collection is reduced, and the solution of the queuing network model is simplified. With this approach, however, the multi-class system is represented by a single-class model. A loss of accuracy results. It has been recently demonstrated that the performance of a multi-class network is bounded below by its single-class counterpart model and is bounded above by a simple function based upon the single-class model. In this paper, a new workload characterization technique is proposed which requires: the K average device demands for the single-class counterpart model and a segregation measure, a value which indicates the degree to which different customers tend to utilize different parts of the network. The segregation measure specifies the point between the two bounds where the multi-class model's performance lies. This measure is quite intuitive and is simple to calculate. The technique provides an accurate estimate of the performance of a multi-class network. 6 refs., 5 figs., 3 tabs.