Implementation And Optimization On The Network Storage System

With the development of the information technology, some requirements for the enterprise storage are put forward, such as higher storage capacity, higher reliability, higher speed and easy access to network. In this paper, a network storage system is discussed and achieved which runs on embedded Linux. Based on the thorough analysis of Linux kernel, the disk controller drivers, RAID, filesystems, the buffer management and the network subsystem are particular expatiated to solve the difficulties in the project.Firstly, the MARVELL 88SX6081 disk controller driver is migrated based on the Linux2.6.10 original SATA driver, which is achieved by the stratification and command transition strategies. Some functions, such as disk monitoring and command queue are added.Secondly, in the implementation of the RAID, this paper makes usage of Logic Volume Manager to overcome the limitation of the flexible RAID subsystem. Comparing the performance of commonly used journaling filesystems using RAID 5, the author solves the problems on fliesystems choosing, and optimizes RAID 5 system in two ways: the first is to improve the compute performance of XOR by Application Accelerator Unit of the processor; the second is to provide a better buffer management solution.Thirdly, based on the multilevel backup cache management, the paper improves the hit ratio by region-LRU and ping-pang buffer management strategies. And the bottleneck when data is destaged to one cache disk can be overcome by distributing the cache partition to every disk. At the same time, cache partition can run in the mirror mode, which enhances the reliability of the data.At last, the data can be copied directly from a disk subsystem to a network subsystem using the primitives like page mapping in the Linux. With page mapping, a page in a kernel address space is mapped into the address space of a user process and the mapped page can be accessed directly by the user process without copying to its address space. So data prefect policy, which combines ACK responses and the transmission, can reduce the usage of the processor and the number of switch among the processes, and also can enhance the network transmission.