CCE Theses and Dissertations

Date of Award

2020

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

College of Computing and Engineering

Advisor

Gregory E. Simco

Committee Member

Francisco J. Mitropoulos

Committee Member

Sumitra Mukherjee

Abstract

Peer-to-peer networking overcomes the single point of failure and bandwidth limitations inherent to the centralized server model of file-sharing. It is both a popular means of sharing digital content and a major consumer of internet traffic, with BitTorrent being the most-used protocol. As such, significant research has gone into improving peer-to-peer performance in order to reduce both download times and networking costs. One aspect that can affect performance is the client’s selection of peers to download from, as the time spent downloading from even a single poor-performing peer can impact the overall download duration.

A recent peer selection strategy explored having a client use historical knowledge acquired through third-party sources, as well as its own first-hand experience with previously visited peers, as a means of selecting likely good-performers, coupled with a peer switching strategy that replaced peers whose post-selection downloads exhibited poor performance contrary to what historical knowledge suggested in order to limit the time spent downloading from said poor-performers Though this tactic demonstrated reduced download times compared to various past works, it still suffered from poor peer selection due to its historical knowledge not necessarily reflecting the current state of the peers.

This work introduced and examined an enhancement to this hybrid peer selection and switching strategy by adding current intelligence regarding a peer’s available bandwidth, all the while avoiding the additional network costs associated with performing on-the-fly probing or querying techniques utilized by other peer selection strategies to benchmark prospective peers. With such on-the-fly knowledge about a peer’s current bandwidth availability, this new enhanced strategy quickly replaced poor performers without waiting for downloads to be performed and subsequently benchmarked, resulting in reduced overall peer-to-peer download times.

The results of adding this pre-download peer switching enhancement demonstrated improved download performance, particularly in early file transfer runs. However, as more runs occurred and the benefits of the original strategy’s historical knowledge became more pronounced, the time savings gained from this new enhancement diminished.

Share

COinS