The Discovery of Self-Similar Traffic

Abstract

With the proliferation of packet-switched networks in the 1980s and a strong preference in parts of the engineering and communications research communities to work on ATM (asynchronous transfer mode) related problems, teletraffic theory, that is, the application of queueing theory to the performance analysis and evaluation of communication networks, faced a new set of challenges, some of which questioned its very foundations. Originally developed to supply engineers with adequate quantitative methods and techniques for designing, managing and controlling circuit switched networks providing plain old telephone service (POTS), teletraffic theory had become arguably one of the most successful applications of mathematical modeling in industry. However, in view of the widespread and rapid deployment of packet switching based technologies and new services, and especially because of the explosive growth of the Internet, traditional teletraffic theory found itself under intense scrutiny. In particular, as some of the arguments went, performance models are only as good as their underlying assumptions - although teletraffic theory had produced a steady supply of new and ever more complex models for packet traffic, the models’ foundations had generally been based on POTS-based tradition, on good faith, or on the modelers’ own (mis)conceptions about the presumed dynamics of traffic in a packet-switched networking environment. Put differently, since traffic modeling is considered to be an essential ingredient of teletraffic theory and since describing packet traffic had become largely a theoretical exercise that was essentially disconnected from reality during the emergence of packet-switched networks, the question of whether or not all these theoretical models had anything in common with measured traffic as observed on links within “live” packet networks could be heard more and more frequently.