BitTorrent Under a Microscope: Towards Static QoS Provision in Dynamic Peer-to-Peer Networks Tom H. Luan*, Xuemin (Sherman) Shen* and Danny H. K. Tsang § * University of Waterloo § Hong Kong University of Science and Technology BitTorrent (BT): A Brief Introduction BT, first appeared in October 2002, is a file distribution system based on the P2P paradigm Engrosses about 30% of all Internet traffic volume [1] Leads to the proliferation of P2P media streaming using the user-driven data-oriented download approach For example, CoolStreaming, PPLive [2] and PPStream for live and on-demand video streaming PPlive is reported in [2] to broadcast to over 200,000 users in one event at the bit rate of 400-800 kbps Successful media streaming requires providing users with the static and guaranteed download throughput [1]. EContentMag.com, “Chasing the user: The revenue streams of 2006”, December 2005 [2]. Xiaojun Hei, Chao Liang, Jian Liang,Yong Liu and Keith W. Ross, "A Measurement Study of a Large-Scale P2P IPTV System", IEEE Transactions on Multimedia, vol. 9, no. 8, pp. 1672 - 1687, Dec. 2007. 2 BT Under a Microscope IWQoS’10 QoS in P2P Content Distribution QoS provisioning is tough in P2P P2P network is inherently dynamic and heterogeneous The heterogeneous bandwidth of peer uploaders results in the unpredictable download throughput of nodes The dynamic nature of peer uploaders results in the intense variance (or jitters) of download throughput to nodes Problem Statement: How to accommodate the bandwidth heterogeneity and dynamics of peers to provision nodes with static and guaranteed download throughput? Methodology: Evaluate and enhance the performance of BT 3 BT Under a Microscope IWQoS’10 BT Protocol BT strives to ensure (proportional) fairness: Nodes attain the download rates proportional to their upload rates Tit-for-Tat scheme (Forbid freeriders) Each node only uploads to others who are uploading to it Choking algorithm (Preserve the high-rate uploaders) Incentive mechanism to encourage the upload Every Tc (e.g., 10) seconds, select nc (e.g, 4) nodes to unchoke (upload to) among the peers which are uploading to it Optimistic unchoke (Explore the high-rate nodes for data exchange) 4 Randomly unchoke no (e.g., 1) node which is not uploading to it every To (e.g, 30) seconds BT Under a Microscope IWQoS’10 Example of the Node Connectivity Download from others via optimistic unchoke of others Upload to others with its optimistic unchoke Data exchange governed by tit-for-tat and choking algorithm 5 Fixed number of upload connections Random number of download connections BT Under a Microscope IWQoS’10 Throughput Analysis of a Random BT Node Assuming two classess of peers, high bandwidth (H-BW) and low bandwidth peers Model the download connections of a randomly tagged node in class as a Markov process with state ( X (t ), Y (t )) Downloading from x H-BW nodes and y L-BW nodes c , c Upload capacity of H-BW and L-BW Download rate at time t H L nodes, respectively. N Asymptotically, the mean and variance of d (t ) are, respectively, Mean population of peers. pH, pL Portion of H-BW and L-BW nodes, respectively. pL 1 pH ( x, y ) Steady state of the Markov process and 6 BT Under a Microscope IWQoS’10 Numerical Solution Transition rates are composed of three events Dynamic node arrivals and departures Connections/disconnections due to the choking algorithm Connections/disconnections due to the optimistic unchoke Obtain the steady state probability with the balance equations where 7 is the transition rate matrix of the node in class BT Under a Microscope IWQoS’10 Model Validation Session level simulator coded in C++ Poisson arrival to the network at the rate of peers/s Mean network size to be N Nodal departure rate / N Each experiment with 30 simulation runs and 95% confidence interval 8 BT Under a Microscope IWQoS’10 Download Rate of Tagged Node over Time Highly dynamic due to peer churns and the frequent disconnection of choking algorithm and optimistic unchoke Download rate is proportional to upload rate 9 BT Under a Microscope IWQoS’10 Increasing nc and no nc: connections in the choking algorithm no: connections in the optimistic unchoke Our model is more accurate to capture the dynamic nature of P2P Fan, B., Chiu, D.-M., and Lui, J. “Stochastic analysis and file Increasing nc improves the fairness Fan: availability enhancement for BT like file sharing systems”, In proc. Increasing no degrades the fairness of IEEE IWQoS, 2006 10 BT Under a Microscope IWQoS’10 Increase Tc and Arrival Rate Tc : Time interval for executing choking algorithm To = 3Tc : Time interval for executing optimistic algorithm Increasing Tc degrades the fairness as nodes are slow to adapt Increase arrival rate degrades the fairness as the network becomes more chaos 11 BT Under a Microscope IWQoS’10 Optimize BT Parameters Given the peer arrival rate and mean network size, we can optimize the parameters of BT towards maximal fairness as Parameters including: number of links and execution frequency for choking algorithm, and those of optimistic unchoke Rather than fine tune the parameters, can we improve the protocol for better performance? 12 Enhanced protocol for better QoS provisioning BT Under a Microscope IWQoS’10 Node Clustering in BT BT relies on node clustering to provision QoS Nodes of similar upload capacity tend to form clusters to exchange data 13 BT Under a Microscope IWQoS’10 Protocol Enhancement What is wrong with the clustering in BT? Optimistic unchoke: blind search Randomly connect to nodes in the peer ocean to explore high rate nodes Choking algorithm: a trail-and-error manner Time to locate appropriate cluster peers is long cluster effect is weak in a highly heterogeneous and dynamic network Random walk based peer selection 14 Efficiently and fast search cluster nodes BT Under a Microscope IWQoS’10 Link Level Homogeneity Form the graph in which nodes have equal capacity per out-degree Make outgoing connections of nodes proportional to their upload capacity With TCP connection, bandwidth is equally allocated to upload connections Random walk algorithm to search peers with high capacity per out-degree value Guaranteed fairness: each connection is bidirectional, downloading and uploading at the same rate 15 BT Under a Microscope IWQoS’10 Simulation A more heterogeneous network with capacity distribution where Download rate of the tagged node over simulation time Enhanced BT with random walk Approaches to the upload capacity with vary small variations in the dynamic network 16 BT Under a Microscope IWQoS’10 Validation of Link-level Homogeneity Over 75% of peers have equal capacity per upload connection, with the value same to the analysis 17 Change the upload capacity of the tagged node every 1000 seconds In practice, upload capacity is shared by multiple applications BT Under a Microscope IWQoS’10 Conclusions To provision static and accurate QoS guarantee is a fundamental and important issue for P2P content distribution networks (e.g., BT, PPStream) We propose a Markov model to evaluate the download rate of a randomly selected BT node How to address the network dynamic and heterogeneity Throughput in the dynamic and heterogeneous network Describe an enhanced BT protocol with efficient peer selection using the random walk algorithm 18 The Blind trial-and-error search is inefficient BT Under a Microscope IWQoS’10 Q&A Thank You ! 19 BT Under a Microscope IWQoS’10
© Copyright 2026 Paperzz