Fuzzy logic based techniques for connection admission control in ATM networks

This paper describes an approach for CAC that uses fuzzy logic techniques and a method for learning the behaviour of the traffic in an ATM link. The learning process is performed by collecting, at specific times, examples of information on the current traffic pattern in the system. Using fuzzy reasoning, the system will then infer, from the knowledge extracted from the examples, a value for the bandwidth required for the new source, when the source is operating within the current traffic pattern. An investigation is made into how the system deals with traffic generated by on-off sources, each with the same peak and mean; the results focus on the statistical multiplexing gain and cell loss. The statistical multiplexing gain increases for small mean to peak ratios (high burstiness)