Research On A High Reliable And Scalable Cluster-based Storage System

With the development of the computer technology and network technology, various kinds of data present the exploding type of geometric series to increase. Various kinds of online data increase rapidly. This impels the storage capacity of the data to increase sharply at the speed of 3 to 5 times of every year. A large amount of application with intensive data, such as the digital library, data warehouse, data mining, meteorological data processing, medical video picture data processing, scientific research of the biology, multimedia requesting, online data processing, etc, put forward higher request to the performance (I/O bandwidth, throughout rate, response time, etc.) to the storage system. How to construct a high performance, high reliable and high scalable storage system becomes an important problem that the storage system faces at present. Aiming at the choke point of storage service in the current network storage technology, a high-reliability scalable PC cluster-based storage system is proposed, as well as a disk striping scheme of the storage nodes in the system. In this scheme, different placement and location techniques are adopted according to the data object scales. The pseudo-random placement algorithm of small data objects is then discussed in stress, which can effectively locate data objects during the system scaling. In addition, the system fault tolerance is achieved by the effective replication of data objects. Finally, the system performance is analyzed.