Defending against Malicious Behaviors in BitTorrent Systems.

摘要

Peer-to-Peer (P2P) systems are a promising and cooperative model for the recent Internet applications, such as content (file) sharing, multimedia streaming, and distributed computing. Unlike traditional client server models, P2P systems are based on participants' active cooperation. This cooperative nature is helping to develop various applications, and contributing toward common goals for their participants. P2P systems can accommodate a great number of participants. Despite scalable service, mutual benefits among participants are achievable through cooperation, without dedicated servers in the systems. In particular, BitTorrent systems, a typical P2P approach, maximize peer cooperation with simple and scalable operations. The BitTorrent systems have not only attracted a great number of users through efficient content distribution, but also have contributed a major portion of the Internet traffic by the scalable users.;Unfortunately, the lack of security mechanisms in P2P systems makes compliant peers vulnerable when achieving peer cooperation. From the early stages of P2P systems, there was little concern about security issues guarding against strategic attackers. Most P2P systems including BitTorrent have optimistically expected users to follow the protocol descriptions, and mainly concentrated on various functions of management for applications. As a direct effect of such conditions, selfish users (freeriders) can take advantage of protocol weakness for their own profits violating the regulated standards. Moreover, malicious users can launch various attacks on compliant users, which will eventually lead to system collapse.;In this dissertation, we investigate vulnerabilities of BitTorrent systems, and propose general solutions to address the problems, such as Sybil attacks, Free-riding, and Byzantine attacks (cheating attacks). These attacks are difficult to prevent due to the typical P2P paradigm with no central authority. Therefore, our solutions are fully decentralized against malicious or selfish participants in P2P systems. Each user makes its own decision without an authority's intervention. Although a peer's decision depends on directly connected neighbors, cooperation with them can improve detection through extension of the local view, and prevention against the malicious attacks. Based on peer cooperation, the proposed approaches defeat selfish behaviors, cheating attacks, and malicious attacks triggered by fake identities. They also help participants achieve performance enhancement in spite of the seriously unfavorable conditions in which various attacks are.;First, we propose GOLF with locality filtering against threats from Sybil attackers in file-sharing systems. GOLF scheme fosters peer cooperation to exclude potential attackers, and locality filtering tentatively identifies Sybil attackers based on patterns in IP addresses. Under the proposed scheme, the effect of Sybil attackers sharply decreases. GOLF effectively detects and blocks potential attackers, despite false accusation.;Second, we present CoVoD, a novel cooperative P2P VoD streaming approach, to penalize uncooperative peers. CoVoD adopts a referral based neighbor management scheme. Each peer expands its local view through referral information, and rewards the source of the referral. Such referral incentives can achieve direct cooperation among peers with asymmetric interests. Experimental results demonstrate that CoVoD improves service quality over alternatives, and strongly penalizes freeriders (peers who do not contribute resources).;Third, we address a defense mechanism, called AntiLiar, against cheating attacks in mesh based streaming systems. It provides participants with a practical authentication and informational integrity based on a secure progress log and a verifiable evidence. A consistent neighbor management improves peer cooperation and blacklists malicious liars. Experimental results show that AntiLiar improves costs to defeat cheating attacks, and achieves a good service quality in a mesh based streaming system.