Performance of adaptive connection admission and flow control with controlled server in ATM network

To analyze the adaptive connection admission and flow control (ACAFC) strategy, we need a queueing model with complex arrival and service characteristics using the underlying arrival phase and service type processes. To solve the queue, we have to study the matrix analytic approach. We use the blocking method (BM) to get the exact solution for the finite queue, however, when the system size becomes larger, the method is slow; thus, we use an infinite queue to approximate the finite queue with the normalizing method (NM). It is shown that the BM is slow compared with the NM based on the adaptive Newton-Kantorovich (ANK) algorithm for an infinite queue, when the system dimension is larger than 35. We investigate how accurate the NM is when it is applied to the matrix analytic model, and the results show that the NM works pretty well as long as the load intensity is less than 0.9; and for most of the common situations we met, both methods consistently show that the level of cell loss probability, for an ATM network using adaptive CAFC, is inversely and linearly proportion to the buffer size. A new call is accepted only when the cell loss probability is less than a certain level, or the quality of service (QoS) is satisfied. The linear relation allows us to obtain the performance of a large system based on the calculation of a small system which is fast. This may provide us with a fast ACAFC algorithm for ATM networks