1. Data networks and data network traffic
In this part of the talk, I will give a general introduction into data networks and data network traffic, and will discuss why modeling large-scale internetworks such as the global Internet is an immensely challenging undertaking. Some of the main problems are discussed in "Why we don't know how to simulate the Internet" by S. Floyd and V. Paxson
2. Modeling network traffic is easy, understanding it is hard ...
The math/stats community has a tendency to use network measurements (in particular, traffic traces) for the sole purpose of ``data fitting'' -- given a trace that exhibits a (statistical) characteristic X, what's the `best-fitting' model that captures property X? While there may be some(?) value to this ``black-box'' modeling approach that has been advocated by traditional time series analysis, I argue that it is ill-suited for the kinds of measurements that are available form today's networks. More importantly, it completely ignores the new challenges posed by the networking application and avoids dealing with fundamentally new issues related to some recently observed "emerging phenomena" in Internet traffic. For a discussion of the relevance of ``black-box'' approach for modeling network traffic, see "Where mathematics meets the Internet" by W. Willinger and V. Paxson