Simulation driven resource allocation in ATM networks

Asynchronous Transfer Mode (ATM) traffic is characterized by a wide variety of traffic source classes. Each class of traffic has differing effects on the buffer and link capacity resources of the network. This results in a high level of complexity in predicting resource allocations for each source. The network must be managed in real-time in order to avoid congestion which tends to collapse its utilization. Network congestion requires admission control which in turn must require a measure of resource requirements from a randomly variable source. In this paper we proposed a frequency-domain probability density function convolution method for predicting the resource requirements of a source based on its input characteristics. A taxonomy and classification of sources based on measurable characteristics is also proposed. The classes of sources are simulated and the resulting requirements are compared to the prediction. A feedback mechanism is described which can result in a highly accurate and very efficient table lookup method for realtime resource allocation in ATM multiplexing switches.