Using peer-to-peer communication to improve the performance of distributed computing on the Internet

This paper presents the design and evaluation of a distributed computing platform on the Internet that employs peer-to-peer communication for transferring data directly among participants without consuming scheduler's bandwidth. The aim is to reduce network congestion at the scheduler by utilizing available peer-to-peer bandwidth. The peer-to-peer model is evaluated and compared to the client/server model in different system settings. The impacts of various factors, including the dynamic nature of a peer-to-peer system, on the performance and scalability are studied. The experimental results show that when peer-to-peer communication is applied, even partially, the bandwidth demands on the scheduler significantly reduced. This results in improved system's performance and scalability, and better support for data-intensive applications.