Can Autonomic and Strategic Routing Benefit Internet Autonomous Systems?

The current complexity of Internet routing has resulted in Internet domain administrators having little insight into how their routing decisions affect other parties and themselves. They often make costly arbitrary routing decisions and learn via trial-and-error. Further challenges to the Internet are posed by the emergence of overlay networks-based user-directed routing which has rendered routing increasingly dynamic. In such environment it is believed that domain administrators and users themselves could find benefit in autonomically analysing each other´s behaviour and acting strategically. This paper finds evidence for this by modelling for the first time Internet domain administrators´ interactions in an environment of user-directed routing by means of Stochastic Game Theory. It introduces a network routing game and assumes this to be repeated over time. At each iteration of this game a domain plays a strategy by advertising a price/penalty to cross their network. End-users then make routing decisions so as to minimise their end-to-end penalty. In this framework, self-interested domains learn a strategy plan based on their current state. Experiments obtained with a simulation implementation of such game suggest that domain administrators can benefit from acting strategically. This allowed them to run their networks more efficiently by achieving on average greater throughput (with equal of fewer drops) both in normal and failure conditions.