Distributed Averaging and Efficient File Sharing on Peer-to-Peer Networks

Citation

Mehyar, Mortada (2007) Distributed Averaging and Efficient File Sharing on Peer-to-Peer Networks. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/Q9EV-S167. https://resolver.caltech.edu/CaltechETD:etd-01102007-010550

Abstract

The work presented in this thesis is mainly divided in two parts. In the first part we study the problem of distributed averaging, which has attracted a lot of interest in the research community in recent years. Our work focuses on the issues of implementing distributed averaging algorithms on peer-to-peer networks such as the Internet. We present algorithms that eliminate the need for global coordination or synchronization, as many other algorithms require, and show mathematical analysis of their convergence.

Discrete-event simulations that verify the theoretical results are presented. We show that the algorithms proposed converge rapidly in practical scenarios. Real-world experiments are also presented to further corroborate these results. We present experiments conducted on the PlanetLab research network. Finally, we present several promising applications of distributed averaging that can be implemented in a wide range of areas of interest.

The second part of this thesis, also related to peer-to-peer networking, is about modelling and understanding peer-to-peer file sharing. The BitTorrent protocol has become one of the most popular peer-to-peer file sharing systems in recent years. Theoretical understanding of the global behavior of BitTorrent and similar peer-to-peer file sharing systems is however not very complete yet. We study a model that requires very simple assumptions yet exhibits a lot structure. We show that it is possible to consider a wide range of performance criteria within the framework, and that the model captures many of the important issues of peer-to-peer file sharing.

We believe the results provide fundamental insights to practical peer-to-peer file sharing systems. We show that many optimization criteria can be studied within our framework. Many new directions of research are also opened up.

Thesis Files