The principle of Maximum Entropy (ME) and the notion of system decomposition are combined towards the creation of an iterative cost-effective approximation algorithm for the performance analysis of packet-switched buffered Banyan Multistage Interconnection Network (MIN) based Asynchronous Transfer Mode (ATM) switch architectures with arbitrary buffer sizes, multiple input/output ports and Repetitive Service (RS) internal blocking.Traffic entering and flowing in the MIN is assumed to be bursty and it is modelled by a Compound Poisson Process (CPP) with geometrically distributed bulk sizes and Generalised Exponential (GE) interarrival times. The GE distribution is also adopted to represent the random nature of the effective service times of packets due to the combined effects of traffic burstiness and RS blocking.Entropy maximisation implies decomposition of the Banyan network into individual building block queues of switching elements, represented by shared buffer cross bars, under revised GE-type interarrival and service times. Each building block queue is analysed in isolation by applying ME techniques and classical queueing theory, subject to marginal mean value constraints, in order to obtain a product form solution for the joint queue length distribution and typical performance metrics of the network.Numerical results are included to validate the credibility of the ME approximation against simulation, define experimental performance bounds and perform a buffer capacity optimisation across the entire network. • Supported by the Engineering and Physical Sciences Research Council (EPSRC), UK, under grant GR/K/67809. D. D. Kouvatsos (ed.), ATM Networks

No abstract

We present an analysis of Asynchronous Transfer Mode {ATM) traffic data collected from the London MAN {Metropolitan Area Network). The traffic is found to have different characteristics depending on its intensity. The correlations between inter-arrival times of sessions for traffic from heavily utilised sites is insignificant at the 5% significance level and a Poisson Point process is thus appropriate as a model. Traffic from low utilisation sites, on the other hand, shows long-memory characteristics with correlation at lag 40 and more. This traffic is modelled firstly using an autoregressive process and then the well known Markov Modulated Poisson Process {MMPP). The burst size distribution for both correlated and uncorrelated traffic shows distinct peaks which corresponds to the fragmentation as data travels through networks with different Maximum Transport Units {MTUs).