Research On Grid File Management Middleware

The grid computing, which is regarded as the most important innovation after the Internet revolution, has been becoming a new hot issue in the computer science area. An important research fields in grid computing is file management which aims to provide a uniform interface to store, access and transfer files in the grid environment with name transparency, location transparency, protocol transparency and time transparency. This dissertation designes and implements a grid file management middleware to solve file management in grid environment. This middleware, which describes and orgnizes gird files in a standard manner, can manage the grid files in centralized way and support a grid client to access a remote grid file randomly. It hides the complexity of file management for other grid entities. The main contents of research on the grid file management middleware are described as followings:(1) To analyze the existing technologies and the process of file management in grid computing. Then to point out the limitations of current technologies in the Open Grid Service Architecture and to summarize the issues to be solved in the grid file management. To give the architecture of the grid file management middleware.(2) To propose a new naming mechanism that consists of Grid File External Name and Grid File Internal Name to identify all files in a grid. Using this naming mechanism, the algorithm to solve the grid file management is designed. The algorithm uses hash function to store file information in a particular node based on the naming mechanism. The corresponding searching algorithm is also designed.(3) To design a Random Access File Service to support remote file access. Using this service, a client can randomly access a remote file. The 3 optimized technologies of data compression, parallel transfer, and buffer adjustment are proposed in order to improve the performance of service. The failure recovery mechanism is also discussed.(4) To implement a grid file management middleware and analyze its performance by experiments. The results show that the middleware system has both high reliabiltiy and high performance in the grid file management.In the last part of dissertation, several issues for further research are gave.