Search

An efficient media streaming method utilizing a globally load balanced overlay network. This method makes use of capacity per out-degree values to construct and maintain an overlay network for media streaming in a Peer-to-Peer environment.

The Internet has become the standard infrastructure for all kinds of communications. At the same time, the emergence of cloud computing and web services has significantly strengthened requirements for the Internet to display greater robustness to disruptions and higher flexibility to support emerging applications. As hardware components are increasingly reliable and capable of handling a growing volume of Internet traffic, routing has not only become the biggest contributor to problems such as performance degradations and failures, but it also affects network flexibility and its ability to provide services. Furthermore, because of growing Internet penetration, routing is playing an ever greater role in shaping the scalability challenges that need to be met to enable the continued expansion of the Internet. The goal of the thesis is, therefore, to explore the role of routing in building a "Sustainable Internet" that can satisfy the future needs of communications along the dimensions of performance, flexibility, robustness, and scalability. The thesis presents a comprehensive approach to building a sustainable Internet from a routing perspective, and solves the following three challenges. The first challenge stems from the fact that IP networks need to support a wide variety of applications with diverse properties and requirements. It is therefore important to understand how routing can be designed and optimized to flexibly support diverse applications in IP networks with performance guarantees. However, flexibility without robustness to disruptions is of limited benefits, because IP networks face the daunting task of handling increasingly diverse traffic that is displaying a growing intolerance to disruptions. As a result, the second challenge is to seek routing solutions for balancing flexibility and robustness in IP networks. Furthermore, IP networks also need routing robustness for their Internet connectivity. Although it can be achieved by multi-homing, this has

In this paper, we propose a simple protocol for building heterogeneous unstructured peer-to-peer (P2P) networks. The protocol consists of two parts-the joining process and the rebuilding process. The basic idea for the joining process is to use a random walk to assist new incoming peers in selecting their suitable neighbors in terms of capacity and connectivity to achieve load-balancing. The rebuilding process specifies bow the nodes should react when they lose links. In particular, we examine two representative schemes, namely the probabilistic-rebuilding scheme and the adaptive-rebuilding scheme. Furthermore, we provide a detailed analysis to investigate our proposed protocol under any heterogenous P2P environment. We prove that the topology structure of the P2P network depends heavily on the node heterogeneity. The analytical results are validated by the simulations. Our framework provides a guideline to engineer and optimize a P2P network in different respects under a heterogeneous environment. The ultimate goal of this paper is to stimulate further research to explore the fundamental issues in heterogeneous P2P networks.

When constructing an unstructured P2P topology, one should consider the application running on top of it in order to achieve a good performance for the whole P2P system. It is not enough only to consider the "topology" objective when forming an overlay such as minimizing a P2P network diameter because it does not achieve a good performance for the whole P2P system. For example, minimizing a P2P network diameter may overload some peers under a flooding search application because of their excess connections. Therefore, one should consider the behavior of the application on top of the topology such that the topology can adapt itself in order to benefit the application. To fulfil this objective, we propose an application-aware topology formation algorithm which can be "tuned" so as to achieve load-balancing for a spectrum of P2P applications. Furthermore, we provide a very detailed analytical model to understand the behavior of our algorithm under any heterogeneous environment. The analytical results are validated by the simulations.

In this thesis, we consider two fundamental questions in the context of un-structured peer-to-peer (P2P) networks. The first question is how to build an unstructured P2P network by tak-ing peer heterogeneity into account. In general, designing a good topology construction algorithm is not a trivial task because there are several chal-lenges. First, the algorithm should be decentralized without heavily relying on a central agent because the P2P networks typically support millions of users online concurrently and the P2P topology may not be completely known to the users. Second, the algorithm should be simple and lightweight in order to keep the overhead to a minimum. Third, the algorithm should be robust under a dynamic environment meaning that the peers can join and leave the network anytime. Therefore, we propose a very simple algorithm for building capacity-aware unstructured P2P networks such that the connectivity of each peer is based on its capacity to achieve load-balancing. The basic idea of the algorithm is to use a random walk to assist peers in selecting their neighbors with a high capacity per connectivity. Furthermore, we provide a comprehen-sive analysis to investigate the behaviors of the proposed protocol under any heterogeneous P2P environment. The analytical results, which are validated by simulations, provide insightful guidelines for heterogeneous overlay network modeling, planning and protocol design. The second question that we consider is how to efficiently search objects in an unstructured P2P network. Congestion and peer heterogeneity make this question more difficult to solve because we have to consider two important QoS metrics in parallel - hit rate and hit delay. However, this consideration has received little attention from the research community. To achieve an efficient search under heterogeneous networks, we propose a congestion-aware search algorithm by integrating congestion control and object discovery functionality so that the sear