Research And Realization On Replica Management In Distributed File System In Large-scale Network Storage

With the coming of information society, the data that produced by companies mount up rapidly, and more and more companies have to depend on these data to keep their business. One hand, the mass of data need more storage capacity to store them; On the other hand, it is necessary that storage system has the characteristic of excellent performance, high reliability, security and so on, because these data is very important for the companies. To provide service of mass storage, it has become a bottleneck in the way of using single old file server. So, a lot of commerce companies and academicians have devoted to solving this problem. Today, there are two ways : Network Attached Storage (NAS)and Storage Area Network(SAN) can solve the problem of large-scalable network storage efficiently, but it is too expensive for general user to purchase such productions. Therefor, it is significant to research a large-scale, high-reliable and excellent performance distributed file system that can manage lots of file servers. This paper firstly introduce the pivotal problems that designing a distributed file system must be confronted with, and that point out their shortcoming in common by analyzing with the currently prevalent distributed file system. Secondly, it expatiate upon the goal that our large-scale network storage distributed file system(LDFS) will achieve. And according to this target ,we establish the system architecture and the parts that compose the whole. Of course, the module of replica management and the selection strategy of storage nodes are made importance in this paper. In the end, we analyse the reliability of system and test it's performance by simulation. The tested results show that LDFS do well in high-reliability and tradeoff of system load. So, we can make conclusion that the management of dynamic replica is effective in reducing response latency experienced by clients and eliminating hot spots among storage nodes. The emphasis of this paper is the research about the selection strategy of storage nodes and the management of dynamic Replicas in distributed file system. In terms of the application occasion, we suppose a new selection strategy of storage nodes. By comparing with the old ones, it has a number of distinct merits , and can satisfy the high-reliability demand of system at the same time. In our LDFS, the management of dynamic replica execute by the client's request distribution algorithm and replica placement algorithm. It can replicate, migrate or delete the replicas of storage objects that are stored in the system among the storage nodes with the change of the practical client's request. The simulation test shows that the management module of dynamic replica make deep effect on the performance, reliability and tradeoff of load in the LDFS.