| The problem of global concurrency control in the multidatabase systems has been widely studied for years. It has been argued that global concurrency control, especially maintaining global serializability, is very difficult to achieve in practical systems. Existing global concurrency control algorithms do not completely guarantee the global serializability in heterogeneous envirionments which are fully autonomous and support replicas. A concurrrency control algorithm for data consistency of global objects under update transactions will be developed in the thesis and its correctness is proven.;The last few years an explosion has been seen in the research and development of distributed file systems. These systems include SUN/NFS, Spirit, Andrew, Amoeba, Eden, Mach, Saguaro, C-Ring, V, and etc. Of these systems, some support a heterogeneous environment (e.g. Mach, Amoeba) or replication (e.g. Eden, Andrew), but none of them provide services for fragmented distributed files, i.e. fragmented files where both fragments and replicas are distributed. We have designed and implemented a novel distributed file system which supports replication and fragmentation in a heterogeneous environment. That is to say, replicas or fragments (the fragment can be replicated) of a file can be distributed at different sites (it may be different local file system at each site). The file operations for distributed files would be provided through the cooperation of the distributed file servers at different locations. For the users, however, the file is a logical unit, the file operations are handled in his view by a logically centralized file server.;Performance studies for exploring the benefits from replication/fragmentation, simulation experiments have been developed and performed. These experiments show clearly that even with under low dominance of local operations the concept of distributed files exhibits a superior performance over the non-fragmented file systems.;Summary, the result of the thesis is that a heterogeneous distributed file system which includes distributed transaction management has been designed and implemented. The features are the following: (1) Transparent access to distributed objects. (2) Maintenance of data consistency under completely decentralized control. (3) The users can access remote objects by using global and local operations. (Abstract shortened with permission of author.). |
