Fast and accurate estimation of ATM quality of service parameters with applications to call admission control

We demonstrate how fast and accurate estimates of quality of service (QOS) parameters can be obtained for an ATM multiplexer queueing model fed by heterogeneous Markovian traffic sources. We achieve this by constructing a simple approximation of the buffer overflow probability: Pr(Q>B)≈De^-θ*B, where θ* is the asymptotic decay rate of the tail of the distribution, and D is a prefactor obtained from the Bahadur-Rao theorem. Upper bounds on other QOS parameters, such as cell loss ratio, mean cell delay and cell delay variance, can all be derived from this estimate. It has been shown that this approximation is much more accurate than the pure effective bandwidth one Pr(Q>B)≈e^-θ*B, which ignores the statistical multiplexing gain due to a large number of sources. One attraction of the proposed method is that the speed of computation of D and θ* is independent of the size of the system. Thus our analytical techniques can be implemented fast enough for real-time administration of admission control in ATM networks