| Along with the bandwidth and popularity growth of the Internet, the scale of the content delivered on the Internet also expanded. Large file downloading and online media streaming will soon, if not already, become the two most resource consuming applications on the network.;In the case of file downloading, the peer-to-peer (P2P) paradigm has been used extensively in recent research literature to improve the system's scalability. One representative example is BitTorrent (BT), which utilizes the upload capacity of participating users with the help of its tit-for-tat mechanism. BT has been used widely in distributing movies, software, various Linux distributions, etc. It is reported that BT is responsible for 35% of all Internet traffic. In the first part of this thesis, we study problems related to the seeding behavior in the current BT protocol and propose enhancements that can (a) discourage free-riding behavior and at the same time (b) improve the performance of other leechers.;In order to aid understanding of BT and possible protocol changes, we propose a simple yet accurate and extensible model of BT in the second part of this thesis. Our model includes (measured) characteristics of the protocol that have previously been left unmodeled. Our model can be used to explore BT parameter setting as well as help design improvements. We validate our model extensively through a rigorous simulation-based study and PlanetLab experiments and show that our model is quite accurate in predicting BT performance.;In addition to file downloading, real-time online media streaming is receiving more and more attention as well. Video content sharing web-sites such as YouTube have gained significant popularity - it is reported that YouTube has 100 million downloads per day. TV broadcasting stations are also beginning to put their programs online for real-time viewing. Unlike file downloading, online media streaming has strict real-time requirements, which translates into more stringent constraints on network resources. Instead of requiring another infrastructure for delivering the streaming traffic, our work focuses on streaming over the current best-effort Internet, which allows systems to be deployed more easily. However, if we use a traditional streaming approach, through a single path, the current best-effort Internet often fails to provide a good quality of service for such applications. To remedy this, we study the use of multiple paths existing in the network to improve the quality of media streaming. In the third part of the thesis, we focus on the load optimization problem in the context of multi-path streaming. Followed by that, we focus on practical issues in designing a multi-path streaming system that is suitable for real-time live applications, e.g., video conferencing. We propose adaptive schemes with the aim of satisfying the typically stringent loss and delay constraints of such applications and, at the mean time, maintaining a low overhead. |
