Systematic Performance Analysis of Bit-Torrent Traffic

Author Name(s): Rupesh C. Jaiswal, Mousami V. Munot, G. S. Mundada, M. P. Turuk, S. D. Lokhande
Author Email: mvmunot@pict.edu

Abstract

The eruption of the World Wide Web as a medium for information dissemination, movies, music and entertainment, has made it extremely important to understand the characteristics of its traffic. Bit-Torrent (BT), communication protocol for peer-to-peer (P2P) file sharing, continues to still remain a dominant source of upstream traffic worldwide. This wide spread popularity of BT has attracted exponentially increasing attention of the researchers in the networking domain. Existing studies on BT systems have reported analytical performance modeling, which includes Markov chain model, Deterministic model, Queuing network model and Fluid flow model. This research study presents exhaustive survey of BT protocol and various mathematical modeling reported for its performance evaluation under transient and steady condition. Precise and apt comparison of the performance parameters in terms of complexity, precision and flexibility is initially summarized in this research. This study further incorporates rigorous experimentation on a huge dataset (BT- data packet traffic) explicitly created for this research to evaluate and analyze other significant parameters like heavy tailedness, self-similarity, autocorrelation, power spectral density and burstiness ( packet inter-arrival time and packet size), which conform the self-similar behavior of BT traffic . Analysis of such parameters is crucial and inevitable during the simulation of the data traces, designing accurate models and is also required by the network provider to design and manage internet traffic efficiently, in order to avoid unnecessary deterioration of quality of service caused by heavy-tailedness. However, such rigorous and wide-ranging analysis of BT traffic, has received meager attention in the reported literature, and is the key contribution of this research study.

Introduction

In today’s era, the use of digital images is increasing [1]. P2P communication, which uses simple strategy of collecting the data from all sources and sharing it to respective peers, has therefore become more popular over conventional client server approach [2,3]. This new protocol brought millions of music admirer together to exchange their own music assets [4,5]. In 2001 Bram Cohen has made the revolution in P2P communication by writing Bit-Torrent (BT) protocol [6,7]. BT protocol has almost replaced all others protocols in P2P network. Today, BT can be credited for a quarter of all upstream Internet traffic in North America, more than any other traffic source[8]. The scenario remains same across the globe. Nowadays most of the social networking sites also use BT network to share hundreds of Megabytes data to all worldwide nodes in a very systematic way[9-11]. BT is a simple protocol that consists of Leeches (peers who download the file but do not share the whole file with the others), Seed or seeder (peers with complete copy of the file), Swarm (network of peers and seeds -peers are downloading the file and simultaneously uploading), .torrent file (a metadata file that contains the address of server), Tracker (a server that handles the BT data sharing process) and Seeding process (uploading the file after completion of downloading)[12,13]. As depicted in fig. 1, two peers, peer1 and peer2 are downloading the same file xyz.torrent. Peer 2 was already in the network and peer1 has downloaded the xyz.torrent file from web browser by using BT client. Once it has .torrent file, it sends request to the tracker for the peer list. Tracker responds it back with the peer list. When peer gets the list, it will form a peer set containing all its neighboring peers from the list. Peer 1starts downloading from this peer set. As peer 2 was already in the network, it starts uploading to peer 1 along with seed. This seed uploads to peer 2 as well. [14]

Conclusion

A substantial survey for evaluating the performance using various analytical modeling techniques for BT P2P file sharing network is presented in this research (Markov chain model, Deterministic model, Queuing network model and Fluid flow model). To evaluate the performance of BT traffic, this research focused on file accessibility parameters, delay for downloading the file, downloading throughput for each peer and service capacity. From the analysis it is observed that deterministic model are more idealized so it is less accurate compared to all other models. Also it is observed, that analysis of Queuing network model is more complex compare to other models. In addition, these analytical methods are compared on the basis of functional regime, complication, precision and flexibility. Further our experimentation with BT traffic demonstrated that it is heavy tailed. Thus, the data packet communication that takes place among BT system shows heavy tailedness. The presence of heavy tailedness is one of the reasons for the presence of self-similarity in BT traffic. The autocorrelation analysis and PSD graphs evidenced the long range dependent behavior. All the above outcomes pointed out that the BT traffic is bursty. An investigation of such parameters is significant and inescapable for the network-administrator / engineer to structure and oversees web traffic proficiently. Such
thorough and wide-going examination of BT traffic, which has received meager attention in the literature is the key commitment of this research study

68 total views, 1 views today

Download PDF File

About the author: admin