NIST Advanced Technology Program

We are at a unique point in the evolution of telecommunications: The amount of data tra c in all U. S. networks for the rst time exceeds the amount of voice tra c. At the same time, recent research conducted at Bellcore demonstrates convincingly that there is a dramatic di erence between the dynamics of voice tra c and the empirically observed behavior of data or multi-media tra c. This di erence requires new fundamental research into the nature of high-speed packet tra c and the subsequent development of novel methods for engineering modern packet networks with a similar degree of quality as is expected of the current voice network. In particular, new engineering approaches are needed for the planning and designing of e cient and reliable networks, for tra c management and control, for networking equipment design, and also for advising the standards and legislative organizations against codi cation of ine cient engineering solutions that would have to be propagated for a long time to assure interoperability and to protect large investments. Telecommunications is the domain of very large industries, characterized by enormous investments, especially in data communications. In 1992 the U. S. market for the telecommunication equipment was $ 52.4 billion, including $ 27.9 billion for the digital packet networks, and the projections for 1997 are $ 103.8 billion and $ 53.3 billion, respectively [1]. The rapid growth of packet communication networks has a tremendous in uence on the evolution of the national economy, and current design decisions will have impact on the development of the advanced National Information Infrastructure (NII) for many years. It is well understood that even a modest percentage of savings due to more accurate network planning and management, or due to better equipment design translates into billions of dollars nationwide. This white paper is based on the experience gained by Bellcore during the past 2 years with an integrated tra c analysis approach, consisting of high volume and high quality tra c data collection from working packet networks, detailed statistical analyses, and extensive performance studies using analytic methods and trace-driven simulations. Results from extensive statistical analyses performed on very long, precise recordings of tra c in local area networks (LANs), Common Channel Signaling Networks (CCSNs) and other networks provide solid evidence that currently used engineering models for high-speed network tra c are utterly unrealistic. In fact, this work has demonstrated that the traditional Poisson paradigm is invalid for today's data tra c pattern's; instead, strikingly di erent fractal, i.e. self-similar tra c patterns have been observed and have been shown to have serious implications for practically all aspects of network engineering. The results of this integrated approach argue strongly for the need to undertake continuous large-scale tra c studies that are based on an extensive database of measured tra c from a large variety of working networks, and for the timely dissemination of the latest ndings and results to the industry. We propose that NIST in partnership with Bellcore and other industry players contribute to (i) the funding of creating and administering a National Network Tra c Database, and (ii) the funding of an associated research program leading to the dissemination of accurate and useful information