In spite of the Internet's phenomenal growth and social impact, many aspects of the collective communication behavior of its users are largely unknown. Understanding the structure and dynamics of the behavioral networks that connect users with each other and with services across the Internet is key to designing the next generation of protocols and applications. We present a characterization of the properties of the Internet's global behavioral network based on data gathered through Indiana University's partnership with Internet2. The structure of this network offers insight into application usage patterns, traffic distributions, trends, and anomalies. In addition, it allows us to derive application networks that reflect the ways in which various network applications are used together. These networks can be used to detect specific applications and their function based entirely on topological features of network traffic without compromising either network performance or user privacy. A test involving traffic data from 15 million distinct hosts shows that the technique can accurately cluster applications into functional categories and identity a set of non-standard or unknown applications, many of which are dedicated to illegal or malicious activity.
Joint work with Mark Meiss and Alex Vespignani.