Distributed Parallel File System The Dpfs The Ssi Technology

The SSI (single system image) technique, which implements two main needs of the system: net transparent and high availability, in the distributed and parallel file system (DPFS), is the key point in the DPFS.In this thesis, an especial memory file system is presented to implement network transparency. Each site builds up a global memory namespace in which the name can transparently point to an object of resource in the system. In other words, the name in the memory namespace can be redirected to any replication (copy) of an object which is distributed in the sites of the system. And physical object location is not available to the custom of the system which provides the transparent access of the file system.Besides, this thesis adopt the following ideas to maintain high availability: (1) duplication mechanism to get resource availability: when some sites in the system are down, the access of the resource in the down site will not be refused and the system can choose an optimal or vice-optimal site to service; (2) log recovery mechanism to get fault tolerance: when a site in the system is down, all the relative sites will log the revision of the data of the same file between the down site and the relative sites and the down site will get the new log to roll back all the revision or copy the file from the relative sites when the down site is upping; (3) message driving data and information consistency mechanism: when the data or the information in a site of the system is revised, the site will notify all the relative site in the system to revise these files by an message model.On the basis of the SSI technique in this thesis, the distributed & parallel file system DPFS, which can realize the active intelligent fault recovery with other modules, has the characters of network transparency and high availability. The techniques of network transparency and active intelligent fault recovery can resolve two theory difficulties: system bottle-neck and system recovery and then realize none-stop system.The balance of this thesis is organized as follows. In Chapter 1 we review the progress of the distributed & parallel and SSI, In Chapter 2 we discuss the basic concept, main needs, key problems, and typical systems of the distributed & parallel file system and SSI. Then, In Chapter 3 we generally discuss the design and implementation of the DPFS, In Chapter 4 we present how to design and implementation the SSI in DPFS, In Chapter 5 several mechanisms of a key problem, consistency, are explicated, In Chapter 6 the availability and performance analysis of the DPFS which adopt the SSI discussed by this thesis are also discussed, The last chapter draws the conclusion and makes expectation for the future research.