CEC Theses and Dissertations

Date of Award

2017

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computer Science (CISD)

Department

College of Engineering and Computing

Advisor

Gregory Simco

Committee Member

Francisco Mitropoulos

Committee Member

Sumitra Mukherjee

Abstract

Peer-to-peer (P2P) applications have become a popular method for obtaining digital content. Recent research has shown that the amount of time spent downloading from a poor performing peer effects the total download duration. Current peer selection strategies attempt to limit the amount of time spent downloading from a poor performing peer, but they do not use both advanced knowledge and service capacity after the connection has been made to aid in peer selection. Advanced knowledge has traditionally been obtained from methods that add additional overhead to the P2P network, such as polling peers for service capacity information, using round trip time techniques to calculate the distance between peers, and by using tracker peers. This work investigated the creation of a new download strategy that replaced the random selection of peers with a method that selects server peers based on historic service capacity and ISP in order to further reduce the amount of time needed to complete a download session. Peer-to-peer (P2P) applications have become a popular method for obtaining digital content. Recent research has shown that the amount of time spent downloading from a poor performing peer effects the total download duration. Current peer selection strategies attempt to limit the amount of time spent downloading from a poor performing peer, but they do not use both advanced knowledge and service capacity after the connection has been made to aid in peer selection. Advanced knowledge has traditionally been obtained from methods that add additional overhead to the P2P network, such as polling peers for service capacity information, using round trip time techniques to calculate the distance between peers, and by using tracker peers. This work investigated the creation of a new download strategy that replaced the random selection of peers with a method that selects server peers based on historic service capacity and ISP in order to further reduce the amount of time needed to complete a download session. The results of this new historic based peer selection strategy have shown that there are benefits in using advanced knowledge to select peers and only replacing the worst performing peers. This new approach showed an average download duration improvement of 16.6% in the single client simulation and an average cross ISP traffic reduction of 55.17% when ISPs were participating in cross ISP throttling. In the multiple clients simulation the new approach showed an average download duration improvement of 53.31% and an average cross ISP traffic reduction of 88.83% when ISPs were participating in cross ISP throttling. This new approach also significantly improved the consistency of the download duration between download sessions allowing for the more accurate prediction of download times.

Share

COinS