A multi-level approach for evaluating internet topology generators

The topology of a network (connectivity of autonomous systems (ASes) or routers) has significant implications on the design of protocols and applications, and on the placement of services and data centers. Researchers and practitioners alike need realistic topologies for their simulation, emulation, and testbed experiments. In this work, we propose a multi-level framework for analyzing Internet topologies and their evolution. Our multi-level framework includes novel measures, evaluation strategies, and techniques for automatically learning a representative set of graph measures. We apply our framework to analyze topologies from two recent topology generators, Orbis and WIT, according to how well they achieve their advertised objectives. The generated topologies are compared to a set of benchmark datasets that approximate different views of the Internet in the data (trace-route measurements), control (BGP), and management (WHOIS) planes. Our results demonstrate key limitations of popular generators, and show that the recent Internet clustering coefficient and average distance are not time-invariant as assumed by many models. Additionally, we develop a taxonomy of topology generators, and identify key challenges in topology modeling.