Abstract
In this thesis, we propose and implement a Distributed Hash Table-Based Metadata Server Cluster (DHT-MDSC), which can provide an efficient routing strategy and an automatic reconfiguration protocol to eliminate the bottleneck of a centralized hash table. The design is based on a new concept merging p2p system with conventional metadata file system into a scalable and high performance distributed metadata server. Furthermore a novel caching mechanism (LC-RIC) is also implemented to improve the performance and scalability of the system. In addition, we have also implemented a reliable DHT-MDSC (RDHT-MDSC) which can tolerate the failure of multiple metadata servers as long as no side-by-side metadata servers fail simultaneously. The system can reconstruct itself upon detecting any fault under normal operation. The time it takes to reconstruct the system is also short in our design.