Internet Traffic Data
William S. Cleveland, Don X. Sun · Journal of the American Statistical Association · 2000
Internet engineering and management depend on an understanding of the characteristics of network traffic. Statistical models are needed that can generate traffic that mimics closely the observed behavior on live Internet wires. Models can be used on their own for some tasks and combined with network simulators for others. But the challenge of model development is immense. Internet traffic data are ferocious. Their statistical properties are complex, databases are very large, Internet network topology is vast, and the engineering mechanism is intricate and introduces feedback into the traffic. Packet header collection and organization of the headers into connection flows yields data rich in information about traffic characteristics and serves as an excellent framework for modeling. Many existing statistical tools and models — especially those for time series, point processes, and marked point process — can be used to describe and model the statistical characteristics, taking into account the structure of the Internet, but new tools and models are needed.