Performance Evaluation of Networks: New Problems and Challenges

Abstract

The purpose of this talk is to give an introduction to the domain of Performance Evaluation of Networks, its methods and its practical results. The short tour will begin with the classical results and finish with some of the principal challenges faced by the theory today. The concern about mathematically predicting the performance of communication systems is not new: the beginning of the theory is traditionally associated with the work of A. E. Erlang (1917) on the blocking probability for telephone trunk lines. The family of stochastic models used by him and his followers eventually led to Queuing Theory, a wealth of formulas and methods for computing throughputs, waiting times, occupation levels of resources and other performance measures. From the point of view of networking, one of the main achievements of this theory is perhaps the family of product form theorems for networks of queues, obtained in the 70’s. When they apply, these theorems allow reduce the analysis of a network to that of each of its elements in isolation. Among numerous possibilities, the results have been applied to the design of scheduling mechanisms for computers, to the problem of resource allocation, in particular the optimal routing in the then-emerging packet switching networks, and to the design of window fiow-control mechanisms. In the 80’s, new problems appeared with the evolution of networking to higher speeds and to the integration of the services offered by classical telecommunications and computer networks. More stress was put on the necessity of an end-to-end “quality of service” (QoS), and “real-time” operation. In parallel, it was realized that the applications that were to use the networks (voice, video, data retrieving, distributed computing) generate a network traffic very different than the usual Poisson processes commonly assumed in the models. All this provoked the emergence of new concepts such as trafic shaping, and the equivalent bandwidth of complex sources. The importance of the scheduling policy for switching nodes in networks has been emphasized. Current research also tries to assess the importance of the long range dependence and fractal behavior of the traffic, which has been measured in local as well as in wide area networks. Even more recently, the popularization of the Web has provoked a renewed interest in the analysis of the performances of the Internet, its protocols, its applications and its evolution. To name just some areas for research: Internet performance & QoS. Flow control & congestion avoidance: TCP and its improvement. Reliable multicast. Feedback-less communication and forward error correction. Differential service. Traffic shaping, policing, pricing. Network interconnection and tunelling. Web performance. Information transfer Protocols, HTTP 1.1 vs HTTP 1.2.Web server optimization: caching, multi-threading, mirroring. Voice & Video. Network-conscious & adaptive compression and transmission. Dimensioning of buffers, playout. Real-time vs offline handling of video on demand. The theoretical foundations of performance evaluation are currently receiving contribution from other fields of applied mathematics: statistics (time series analysis, parameter estimation), optimal control theory, game theory (fairness of resource sharing, individually vs socially optimal behavior).