Research On Optimization Of P2P System Based On Behavior Measurement

With continuous improvement of PC performance and rapid development ofnetwork bandwidth, lots of new distributed applications are emerging in the Internet.Peer-to-Peer network (P2P) is undoubtedly one of the most successful andwidelyusing applications in early21st century. Different from traditionalclient/server model, P2P adheres to design philosophy in early period of Internetdevelopment, takes full advantage of mass resources (computation, storage,bandwidth, etc) on the edge of network, forms a huge and scalable overlay networkand provides good services to the users. The convenient, economical and elasticresources in P2P are attracting more and more attentions from the enterprises andusers. It is becoming the focus of the academia. However, similar to the Internet,the vast scale of P2P network poses an obstacle to the direct evaluation andimprovement itself without profound understanding. Inappropriate design andincomplete security mechanisms not only result in the loss of performance, but giveattackers the opportunity to exploit P2P for attacking, which are confining itspopularization and development severely. This dissertation is dedicated to profilingthe behaviors of P2P network by means of network measurement in order tooptimize P2P system.Firstly, this dissertation systematically introduces the concept, characteristic,architecture and classification of P2P network, and presents a detail survey ofresearch on P2P measurement and security issues. Then, current popular P2Pnetworks are studied and measured in the following views separately: systematicbehavior, external behavior, internal behavior and attack behavior, and severalimprovements and suggestions are proposed specifically.Because the overall behaviors of P2P network are changing gradually with thedevelopment of P2P technologies, existing measurement methods cannot carry outall aspects of P2P systems, whose efficiency and integrity need to be promoted.Since BitTorrent is a kind of hybrid network with complex components crossingmultiple networks, the study on it can serve an effective guide for the measurementof other P2P systems. This dissertation proposes an efficient measurement system for BitTorrent, including torrents collection, peers probing, pieces gathering andpassive monitoring. Then based on the system, a complete and representativemeasurement is performed for a long time to profile the characteristics of content,core components, peers and pieces, and point out potential performance bottleneckor influencing factors. Furthermore, an improved model is also proposed to depictthe daily pattern of peer arrival rate which is not well modeled before, and tworesources with different popularity are also contrasted.The macroscopic behaviors mainly stand for group characteristics of allparticipants, but individual characteristics need more elaborate measurement andobservation. The study on external behaviors of single peer provides a direct way toshed light on the internal working mechanisms. In order to investigate the externalbehaviors, this dissertation works on a typical kind of traffic in P2P networks—Distributed Hash Table. DHT traffic identification based on compound behaviors isproposed to describe the distinctive traffic pattern and can identify P2P hostsindirectly. It analyzes the DHT intrinsic mechanisms, makes good use of the routingrole and discovers several behavior patterns, such as bootstrap behavior, routingbehavior, diverse behavior and short-session behavior. The experimental resultsshow this method can distinguish P2P hosts from others.The external behaviors reflect the characteristics in the way of black box whenP2P hosts connect to others, which are the embodiment of internal implement, sothe measurement of internal behaviors can guide system design and improvementbetter. Therefore, this dissertation focuses on the performance enhancement of DHTrouting by optimizing the structure of routing tables and improving the updatingalgorithm. To evaluate the performance, several characteristics of local routingtable are studied intensively, including the size of routing table, growth and theratio of churn, and moreover the effect of different key running parameters(find_self action interval, message timeout and parallel degree) are discussed. K-bucket with mutable size and updating algorithm based on dispersity of K-bucketare proposed subsequently. The results of corresponding experiments show theimproved routing table is better than the original one.The study on attacks is an important aspect of P2P behaviors research, because attacks have a serious impact on sharing efficiency in P2P network. For this reason,this dissertation presents a study on the common attacks in BitTorrent, and analyzesthe principles of different attacks. On this condition, this dissertation analyzes theleechers attack theoretically, and discusses the extra factors to affect the result offake-block attack and points out the design of security mechanism to counteragainst these attacks. The experiments on WAN show that BitTorrent system isvulnerable to leechers and fake-block attacks, which also proves the effectivenessof the proposed defense strategies.Overall, this dissertation systematically researches the behaviors of P2Pnetwork, and analyzes the factors and security threats impairing the performance ofP2P network. The proposed methods and technologies can effectively improve theperformance and enhance the security of P2P network, which is significant for thefuture research on P2P network.