Research On Architecture And I/O Accelerating Strategy In Two-dimensioned Disk Matrix

The fast developing of modern science and technology and the popularizing of internet have given rise to an explosive increase for the information resources of the internet.Jim Gray,the winner of the Turing Prize of 1998,proposed the famous "Moore's Law" of storage field:The increased new information per 18 months equals to all information produced before. As the development of the society, lots of infor-mation needs to be coded,stored and accessed,which require the high performance of the capacity and speed of the information storage system.In a massive storage system that has multiple array controlers which each has multiple disks,in order to achieve the goal of meeting the user's require of capacity and response time,the array virtual engine must has the capability of aggregating the physical storage of the system as one or more massive virtual disk and provinding the speed of answering the main frame's I/O request as quickly as possible.The I/O requests of the storage server of different applications has two sorts of different features:big data size and adjacent address,small data size and non-adjacent address.Based on these state of affairs,this paper proposed two sorts of different vir-tual disk matrix structure to achieve the goal of aggregating the physical capacity:①to direct towards the I/O request of big data size and adjacent address, proposed a disk matrix of single I/O queue.②to direct towards the I/O request of small data size and non-adjacent address,proposed a disk matrix of multiple I/O queues.In addi-tion,this paper analyzed the data partitioning and placement strategy,logical space management and address transformation algorithm of the two matrixes.Based on the matrix structure,this paper proposed the full concurrency access strategy and parallel cache management method of the matrix,which achieved the goal of fast I/O access of multiple clients.At last,this paper proposed the I/O service model to evaluate the performance matrix according to the queuing theory.The theoretical analyses and testing results indicate that not only the matrix can present the host with massive logical volume,but also enormously increased the speed of I/O concurrency access of multiple clients.