Aemen Lodhi - Employing agent-based models to study Internet interdomain network formation, traffic flow and economics

The Internet is a complex network of thousands of interconnected Autonomous Systems or ASes. These ASes act independently, selfishly and often myopically to optimize their fitness. However, their connectivity decisions are not isolated. Complex interconnectivity patterns in the Internet lead to network effects where decisions by one AS have an impact on other ASes and thus influences their decisions and so forth. How can we understand the macro behavior of the entire AS population? How can we analyze the large-scale co-evolutionary network that is the Internet where its constituent ASes affect the network through their connectivity decisions and the network in turn influences their connectivity policies? While not entirely new, agent-based modeling has not yet been widely deployed to answer the above questions. We present GENESIS, a computational agent-based model of interdomain network formation, traffic flow and economics. We also present an application of GENESIS to study the recent wide-scale adoption of Open peering among ASes. Through this talk we also aim to understand the strengths and weaknesses of agent-based models as compared to game-theoretic and analytical approaches. How can we produce more realistic yet computationally feasible models? How can we validate such models? Are there questions which cannot be answered through this approach? Are there questions which can be answered through this approach better than other approaches? What are the sources of error in agent-based models of the Internet? We explore these questions in the talk through GENESIS.