Optimal load and resource balance in internet distributed systems

Graduation date: 2012A distributed system is a network of multiple autonomous computational nodes designed\ud primarily for performance scalability and robustness. The performance of a distributed system\ud depends critically on how tasks and resources are distributed among the nodes. Thus, a main\ud thrust in distributed system research is to design schemes for distributing computational loads\ud and resource sharing in ways that boost the overall system performance. For many Internet\ud distributed systems such as Peer-to-Peer (P2P) networks, the optimization of loads and\ud resource sharing become even more important due to the higher costs of communication and\ud management of nodes located geographically far part. Furthermore, in many P2P-based\ud systems, nodes are individual users who ultimately decide whether they want to contribute\ud their computer resources to the system. As a result, in addition to the system performance\ud consideration from the engineering perspective, many recently proposed resource and task\ud distribution schemes are designed to provide incentives to users, in such a way to promote\ud resource sharing and enhancing system performance.\ud In this dissertation, we study different approaches for optimizing performance of distributed\ud systems. Specifically, the dissertation is focused on "balancing" resources and loads on each\ud node to improve the performance for two important models of distributed systems.\ud In the first model, we study incentive mechanisms to promote cooperation of users in P2P\ud networks. In a typical file sharing P2P network, there are uncooperative peers, or free riders who only download data from other peers without sharing their data with others. As a result,\ud upload bandwidth of cooperative peers are saturated, but those of uncooperative peers are\ud unused. This leads to overall low system utilization and performance degradation. This\ud dissertation proposes a scalable, decentralized mechanism to efficiently prevent free riding, and\ud to increase system performance. The key ingredient of the proposed mechanism is to balance\ud the amounts of upload and download data of each peer using the notion of global contribution\ud which can be efficiently calculated in a distributed manner.\ud In the second model, we consider a class of distributed server systems that can be used as the\ud underlying engines for many Internet applications, especially social networking applications.\ud Such a system consists of a large number of clients who communicate with each other indirectly\ud via a number of intermediate servers. Optimizing the overall performance of such a system then\ud can be formulated as a client-server assignment problem whose aim is to assign the clients to\ud the servers in such a way to satisfy some pre-specified requirements on the communication cost\ud and load balancing. We propose to solve this client-server assignment problem via relaxed\ud convex optimization techniques. The simulation results indicate that the proposed approach\ud produces superior performance than many other popular heuristics