BitTorrent Discovery and Performance Enhancement using DDS QoS Policies

摘要

BitTorrent is the most worldly adopted peer to peer (P2P) file distribution application protocol that constitutes a huge part of today's Internet traffic. P2P model benefits BitTorrent's peers in file exchanging process, by eliminating a single point of congestion. Beside the main P2P implementation of BitTorrent for file exchanging, it also has client-server communication between the peers and tracker. The tracker is used in BitTorrent network for peers' discovery. However, the tracker does not benefit from P2P characteristics. Mainly, the tracker is considered to be a single point of failure, also, scalability and load-balancing are other traker's issues. Another problem is that BitTorrent uses eleven overhead messages which help in distributing the file pieces among the peers. In this research work, we aim to have a pure P2P BitTorrent application, and minimize the messages overhead. For Discovery, we propose a novel architecture to decentralize the tracker and make it distributed among the peers. The proposed method reduce both communication overhead and node searching complexities to O(1). For Dissemination, we re-implement the existing BitTorrent using Data Distribution Service (DDS), which is a Real Time Publish Subscribe (RTPS) middleware. We study the performance in terms of file downloading time and goodput for both the original and the proposed dissemination protocols. The results show that the proposed solution can minimize the BitTorrent overhead achieving high goodput, and speed up the file downloading process in most cases. The proposed approaches are tested and validated only over Intranet.