A PERFORMANCE-MODEL OF THE CREDIT MANAGER ALGORITHM

This paper presents a performance model of the credit manager algorithm that has been proposed for access control for the Switched Multi-megabit Data Service (SMDS), which is a service concept for very high-speed data services. We will present a flexible model that makes few assumptions about the process that describes the requests for access, and allows several types of requests to be considered distinctly (e.g., typing, screen updates, and data transfers). The purpose of the model is to predict the probability that a frame will be lost when the arrival parameters and the credit manager parameters are specified. In general, the model is solved numerically. Some numerical examples, using estimates of anticipated applications, are described. When a bogey for the loss probability is given, the optimal value of credit budget is obtained. Some graphs that show how some traffic parameters influence the setting of the credit budget are presented. Our numerical results show that the stochastic nature of the arrivals can have a large impact on the throughput that can be achieved with a tolerable loss probability. For a fixed value of the credit budget, a user may adapt to the credit manager by storing requests in a buffer. We describe how the buffer decreases the proportion of lost requests, and show how large a buffer is required to achieve a tolerable blocking probability.