Performability Modelling and Evaluation in Peer to Peer Content Delivery Systems under Attacks
MetadataVis full innførsel
Since peer to peer (P2P) content delivery was born in 1999, it has become an important part of the Internet activity. It causes a substantial part of the traffic and involves a large number of participants. Instead of delivering service from central servers, in P2P content delivery systems, peers at the edges of the Internet are collaborating on the exchange content. The integrated service capabilities and the redundancy of the content yield scalability and robustness. This thesis introduces the concept “performability” into evaluating P2P content delivery systems, which concerns their capability of providing desirable service. It reveals the impacts from inherent problems which hamper P2P content delivery systems, like unpredictable peer behavior, malicious peers attacking from the inside, copyright infringement issues and etc. Their influences are analyzed quantitatively in terms of: the QoS an individual user can expect, the distribution of a piece of content in the network, and the energy consumed. Methods to improve the performability of P2P content delivery systems are compared and suggested. Because of their largeness and worldwide distribution of peer locations, it is hard and costly to gain insight into the dynamics of P2P content delivery systems through monitoring. Hence, modeling methods which are able to study time evolution are used. Markov models, stochastic activity networks, and fluid models are selected for making approximations of P2P content delivery systems. In addition, game theory is applied for analyzing the strategies of different participants involved in a same system. According to our knowledge, it is the first time to apply stochastic activity networks in modeling P2P systems. It is also the first time that game theory is used in studying P2P content pollution. Validating simulations, which include more details, are carried out as well. Two modeling approaches are used. One of them is introduced by this thesis. It models a P2P content delivery system from the perspective of an individual user. It can reduce the modeling complexity when many details need to be considered. The other approach which models the evolution of an entire system is more mature and conventionally used. By comparing conclusions drawn by these two approaches, it is illustrated that our novel approach can provide solid conclusions as the conventional one. This thesis provides interesting results concerning performability of P2P content delivery systems. However, how to use these results to benefit ordinary P2P users, copyright owners, and internet service providers is still a challenge in our future work.