Abstract:Peer-to-peer (P2P) emerges as a major paradigm for scalable network application design. It also exposes a fundamental issue in the traditional Internet architecture: emerging applications can have tremendous flexibility in how data is communicated, but if applications are to participate in network resource optimizations, then the networks cannot continue to be opaque but need to provide a communication channel for collaborative traffic control.In this thesis, I propose a framework called P4P to enable cooperative traffic control between applications and network providers. The framework provides multiple privacy-preserving interfaces to allow network providers and P2P applications to jointly optimize their respective performance. At the core of this framework is the p4p distance interface, derived from the principle of primal-dual decomposition. Using this interface, neither do the providers need to know the specifics of applications, nor do applications need to know the specifics of providers or other applications sharing the same network. The framework is fully implemented. The results obtained from both controlled experiments running on the Planet Lab and large-scale Internet experiments running on large Internet service providers including Verizon Communications and the Telefonica Group demonstrate that P4P either improves or maintains the performance of native P2P applications, and that it substantially reduces network provider cost compared with either the native or localized P2P applications.

Explicit Communications for Cooperative Internet Traffic Control