| The aim of Storage Tank, a research project conducted at IBM's Almaden Research Center, was to design and build a first-of-its-kind, SAN-based file system. Some of the important attributes of this file system include direct SAN-based access to data blocks from clients; a true heterogeneous architecture with support for various client platforms; a scalable and highly-reliable cluster of metadata servers, scalability to very large environments; and single system consistency guarantees and locking semantics. A particular focus in designing this system was support of policy-based data lifecycle management, which allows automated control of every aspect of data management throughout a file's useful lifetime.;This dissertation describes the genesis and background of the Storage Tank project, and discusses related work in the area of distributed file systems. It then describes the architecture and operation of the system in some depth. Following those sections is a detailed discussion of the Policy-Based Data Lifecycle Management features of the system. Several companion projects are presented, followed by laboratory and pilot installation experiences with the system. Finally, proposed future work and a discussion of the impact of the system are presented. |
