by Jorma Virtamo
Researchers at VTT Information Technology are studying several problems related to the performance and design of broadband networks, ATM networks in particular, in international collaborative projects. The work ranges from the analysis of basic queueing problems to the traffic characterization of self-similar data traffic and their application on real networks.
VTT Information Technology has been actively involved in pan-European research projects on performance analysis and design of broadband networks since 1986. These projects include three consecutive COST projects, of which the latest, COST242, continues until May 1996,and the Race COMBINE project.
Studies of queueing problems in ATM networks
Queueing theory is a basic element in the performance analysis of all communications networks. Within the COST projects, a close co-operation with the researchers at France Telecom's research laboratories CNET, Paris, led to revitalization of the so-called Benes method, originally presented in the early 60's. This method provides a general approach to queueing problems.
Though the Benes method does not directly give an answer to the queue length distribution but rather represents a reformulation of the problem, it often allows one to derive, by some extra steps, an explicit solution or at least a tight upper bound for the distribution function. By the generality of the method, it can be applied both for systems with point arrival processes and for systems with gradual fluid flow type input process.
In the context of the queueing problems arising in ATM networks, the Benes method has proven to be a powerful tool. Many systems have been successfully analysed with this method, including the N*D/D/1 and SumD_i/D/1 queues, and the fluid queue with an Ornstein-Uhlenbeck input process. Some recentwork has concentrated on the characterization of the output processes from the queueing systems in order to facilitate the study of systems with several cascaded stages of multiplexers.
Optimal routing in networks
Methods of stochastic optimization and Markov decision processes have been applied in the solution of some optimization problems of scheduling systems operated on reservation basis as well as in finding the optimal control policy of a complex multiserver batch service queueing system. The same methods are currently being used to the studies of optimal routing problems in networks, for instance, to the problem of choosing the route for an offered ATM connection belonging to a given class in such a way that the expected benefit (revenue) for the network operator is maximized in the infinite time horizon.
An important ingredient in performance analysis is the traffic characterization. Models that capture the salient features of the real life traffic processes are needed. Recent findings based on extensive measurements of data traffic in Ethernet networks, performed in Bellcore, USA, have shown that the use of the traditional Poisson type traffic models may lead to grossly erroneous results. In fact, the real data traffic exhibits self-similar behaviour with correlations at all time scales.
In a work done at VTT a model based on fractional Brownian motion (FBM) hasbeen proposed to describe this kind of behaviour. The queueing problem with FBM as the input process has also been studied as well as the problem of traffic forecasting on the basis of the known history. The importance of this kind of research for practical applications is rapidly gaining recognition.
In the Race COMBINE project the focus is on the traffic problems and end-to-end performance of interconnected high-speed networks such as ATM and DQDB networks. In such a setting one network, loaded by its internal traffic, may be viewed as a variable rate stochastic server for the other network. The COMBINE project will end at the end of 1994. VTT Information Technology is prepared to participate in new partnerships in the future EU research activities, e.g. in the forthcoming ACTS programme.