P2P Networking Theory

Project Overview

The fundamental premise of this project is that peer-to-peer networking will revolutionize the Internet and computing over the next ten years.  Our vision is that P2P will transform the Internet from a shared bandwidth infrastructure into a combined bandwidth/storage/computing infrastructure, creating a massive, global virtual computer which will be shared by users around the world.

There are already many hugely successful, large-scale deployments of unstructured P2P file sharing systems in the Internet. In parallel, several teams of computer science systems researchers are exploring structured P2P substrates, which promise to support a rich array of new networking applications.

However, P2P systems are enormously complex and create a networking environment that is significantly different from the traditional client/server environment. Specifically, P2P nodes -- which can store, retrieve, serve, forward, search for, and process information -- differ from the traditional model along the following attributes:

• scale:  there is a greater abundance of P2P nodes (expected to reach into the billions);
   
• dynamics: under normal operation, P2P nodes continually join and leave the network;
   
• heterogeneity: P2P nodes have vastly different bandwidth and storage capabilities from one another.

Given the enormous scale and complexity of P2P systems, it is difficult to evaluate new design proposals through online experimentation or large-scale simulation. Indeed, online experimentation of a new P2P design not only requires large-scale deployment, but also requires users to elect to use the new system and permit monitoring of their actions. And large-scale simulation is either costly or altogether infeasible for problems of such scale and complexity. Thus, to forecast the success of new P2P designs, there is a critical need for P2P networking theory, that is,

• scalable, distributed algorithms that are suitable for these large, dynamic, heterogeneous environments;
   
• performance modeling and optimization tools that capture and explain the behavior and limitations of P2P distributed algorithms.

This is a joint project between Polytechnic University, Columbia University and the University of Massachusetts. The project organizes informal, bi-monthly workshops, bringing together the project's principal investigators as well as all postdocs, Ph.D. students and master's students associated with the project. Many of the workshops will also include researchers from nearby industrial labs, AT&T Research, Lucent Bell Laboratories, and IBM Research.