Determination of effective bandwidth in ATM networks

As part of connection admission control, it is necessary to estimate the aggregate capacity required for both new and existing connections. Two techniques of capacity estimation are summarized in the article: large deviations theory (Chernoff bound) and eigenvalue/eigenvector technique. The large deviations theory provides an estimation of the tail distribution for the link saturation probability. For a bufferless multiplexer, the saturation probability can be used as a measure of cell losses. With a buffer, an asymptotic behavior of the multiplexer can be derived for a large buffer size. Usually, the dominant eigenvalue plays a major role and the aggregate capacity of connections can be approximated from the dominant eigenvalue. Numerical results of both techniques are compared in terms of the largest number of connections for homogeneous on/off sources.<>