| Hi-index | 0.00 |
Content distribution networks are the dominant technology for distributing shared media on today's Internet. There are several types of content for which these systems have been proven highly successful: static databases, streaming media, online gaining. At the same time this architecture is not appropriate for other types of applications such as transactional databases that require both strong consistency and wide area replication. One scalable technique that can cope with these requirements is to deploy quorum systems on wide area networks. The most important drawback of this approach comes from the performance penalties that one has to pay when accessing data over the wide area. This thesis makes several steps towards addressing this performance problem. First, it presents approximation algorithms for deploying quorum systems to approximately minimize network delay or network congestion. Second, it presents several heuristics for balancing the impact server load and network delay have on client response times and overall throughput. Third, it shows that the techniques presented here can be effectively used to improve performance through both simulation and emulation of real quorum systems protocols on several wide area network topologies.