Heterogeneous Machine Based On The San Shared File System Design And Implementation

Calculate enviroment in large scale corporations is formed by many dissimilar servers including Open Systems and Mainframes.If calculate resource of Open Systems and Mainframes can be stored centralizedly and accessed sharedly, the storage management cost will be reduced obviously.SAN-based Shared File System for Dissimilar Systems is designed just for realizing shared access of resource between Open Systems and Mainframes. The system has two networks: one is SAN (Storage Area Network), and the other is LAN (Local Area Network). Open Systems and Mainframes content with storage devices through SAN. In virtue of the direct access faculty of SAN, the system can have the great I/O performance closed to local file system. And Open Systems and Mainframes content with each other through LAN which used for transfering the file control information between the both. The control information can avoid shared access conflict between Open Systems and Mainframes. Besides, the file sytem has two key modules: metadate agent module and file access module, which are implemented on Open Systems and used for simulating the Mainframe file system. Through the modules, the system masks the file system difference between Open Systems and Mainframes. And therefore the system can store and access data transparently.In the thesis, all the work and research result can be summarized as follows:1) Many disk access technologies have been researched and used integratively in file access, such as Cache, Prefetching and Write-Clustering, Asynchronism I/O, Raw Disk Access and so on. And according to the space layout of the Mainfram file system, a strategy of adusuting cache size, called BPB/TPB, is proposed. The test results show that by utilizing the strategy, the performance of read and write may be increased at most by a factor of 0.15 and 0.25 respectively.2) Metadate server cluster and redundant I/O paths technologies have been researched. Simple and effective load-in-blance and fault tolerance algorithms, which are suitable for the system, are implemented. For example, used the Mainframe service performance as a weight, a new weighted AP number least algorithms is proposed for the Mainframe cluster's load-in-blance. After using the two technologies, the system's process speed has been improved highly and the system has avoided the single point error, and therefor the system has the high availability and reliability.