The Research And Implementation Of Metadata-manage Cluster's Key Technology In Object-based File System

In information ages, data storage occupies a very important position. Storage has been going to be an important factor to the survival and development of one enterprise. It's a significant task of how to build up a storage system which has high performance, good expansibility and good usability and is easy to manage and safe.Object-based Storage Technology is a new technology in storage field and gives new storage architecture. Object is the kernel of this architecture, encapsulating metadata and data which are managed by respective system. Metadata is managed by metadata server which includes the attributes and access authority of file. File data is striped and stored on the object-based storage devices (OSD). Client file system provides interfaces of storage system to users, can interact with metadata manage system and exchange data with OSDs. The large scale distributed file system established in Object-based Storage Architecture can provide good scalability, high performance and strong data process capability in parallel.Our project focuses on metadata management of object-based storage architecture. The scalability and high performance of metadata service is crucial to the architecture. Manging metadata by cluster is one trend of metadata management. In this thesis, adopt a novel architecture to realize MMC (Metadata Manage Cluster) which manages metadata in hiberarchy. In MMC two clusters of DPISs and MDSs manage directory metadata and file metadata respectively. And there is a research to the key techniques of it.On the basic of research to load balancing, we design and realize the scheme of load-balancing which combines the static load distributing and dynamic redistributing. Metadata load is diffluent or with distributing metadata by static metadata partition algorithm. Servers feedback the load information dynamically and the load that is not balancing is redistributed. MMC is load balancing with the method and even as access "hotspot" data.In addition, a research is done to the usability of MMC. DPISs cluster adopts sharing and tolerating array of disks and the mechanism of detecting failed node to resolve fault of the sharing data and nodes. MDSs cluster can take over the task of metadata service and rebuild backup data by using backup servers and spare server. MMC can tolerate failure and recovery from single failure and multi-failure in special conditions. So the system has high reliability and good usability.