Research And Implementation Of Data Management Technique In Paralell Network File System

As information technology swiftly developed, the increase of information presented an explosive trend, which brought huge challenges to store and manage massive data. Peta-scale data storage became the principal problem calling for urgent solutions, and need organize data effectively to reply to thousands of user requests. All of the requirements from large applications on performance, scalability, security and intelligent data management were beyond the bounds of the traditional storage architectures.New-style storage architecture is one of the available approaches to resolve the above problems. Related works pay attention to integrate NAS and SAN, and replace the traditional file-based and block-based storage by object-based storage, which promises higher performance, higher scalability, higher security and more intelligent storage services. Besides, separating the metadata operations from the data path and dealing with metadata transactions by special servers allowed the clients to access the storage devices directly and fleetly.The thesis lucubrated the Network File System and extended it for parallel prototype which we called SN-pNFS(scalable NAS-parallel Network File System). This prototype realized the separation of metadata and data servers and implemented a file-based storage protocol, by which the prototype eliminated the underlying bottleneck of I/O performance caused by the concentrated server, and provided clients the ability to visit file data simultaneously. Based on it, we introduced the idea of object-base storage, and brought forward an object mapping and placement arithmeticn, named SAOMA. What's more, we synthetized and compared the typical cluster mechanism of metadata management. After that, we integrated both the above two into the parallel framework of NFS and partially completed the design of an object-base storage protocal and data management module.