A capacity allocation rule for ATM networks

One of the key issues is how an ATM network estimates the amount of capacity required for supporting a particular application. In this paper, we consider an ATM network as a flexible multi-rate "circuit switched" network with ATM transport, and identify each application (i.e., traffic source) by a metric, namely its equivalent capacity. The objective of this paper is to determine a rule for estimating the equivalent capacity of the application from its traffic descriptor. We model the cell stream of an application its a point process source, and use the Shannon noiseless coding theorem and entropy rate of point processes to derive its equivalent capacity. The resulting rule shows that the equivalent capacity of an application only depends on its peak rate and mean rate. Furthermore, the rule indicates that (i) the equivalent capacity of an application is linearly dependent on the logarithm of its burstiness, and (ii) it is more efficient to multiplex different streams of a multimedia application at higher layers. We also analyze the performance of a network node that operates according to the capacity allocation rule and supports a number of identical bursty applications to obtain its cell loss ratio. The numerical results with bursty data applications indicate that the node satisfies the QOS requirements of stringent bursty applications, and the rule provides a relatively conservative estimate of the required capacity of an application.<<ETX>>