Utilizing Traffic & User Behavior Patterns in Peer-to-Peer Networks

Peer-to-Peer (P2P) networks gained their reputation throughout the last decade thanks for breaking down communication barriers while enabling media content exchange in a click of a button. Yet, the complexity of P2P architecture had an unfortunate side-effect in the form of fluctuations in download rate and service fairness issues for the users while becoming a heavy financial burden for Internet service providers. This work contribution is in increasing the level of understanding of P2P behavioral aspects and presenting novel solutions for key challenges by utilizing traffic, user and application behavior patterns in P2P networks. The first part of this thesis present how modification of traffic behavior patterns can help service providers with localizing popular P2P networks including encrypted ones, without raising the legal disputes of caching solutions or compromising the QoS as in traffic shaping solutions. Further, the traffic behavior patterns of P2P file sharing networks are being modified in order to gain stable download rates and allowing existing P2P networks to efficiently utilize extremely weak or unstable sources. The second part addresses the effects of user behavior on the performance of P2P file sharing and streaming networks. Using supervised machine learning algorithms the solution presented can learn the behavior patterns of the user and predict the effects of the user on the throughput of his machine, increasing the download rate of P2P file sharing networks and preventing hiccups in P2P streaming networks. The final part discusses how application behavior patterns can help with predicting excessive resource consumption of applications. Unsupervised machine learning algorithms are being used to learn the behavior patterns of each application by utilizing network, processor and file system events. Finally, a solution for identifying encrypted protocols such as Skype behaviorally is presented, without using traditional port or payload based deep packet inspection techniques.