Optimization Research On Reading And Writing Performance Of RAID5

With the rapid development of the computer technology, the access speed of diskstorage devices have been far from matching the CPU processing speed. In order to reducethe difference in speed, currently the number of large-scale server’s disk has been usingRAID technology. To a certain extent RAID5disk array reduce the I/O request responsetime, and improve I/O performance of storage system. How to improve and optimize theread and write performance of RAID5system so as to maximize the effectiveness in thestorage subsystem is always the focus of attention.Currently the I/O storage subsystem uses their own caching system, which exist thephenomenon of duplicate copies in memory, thus leading to an increase in CPU overheadand reducing I/O performance of the disk array. In order to reduce unnecessary memorycopy, we use zero-copy strategy to optimize the read performance of RAID5system.Dealing with I/O read request in this method, the system bypass the kernel cache providedin the driver layer, direct exchange data to improve the system performance. Tests showthat read request in one stripe increase significantly in the improved RAID5system.In a multi-user small data access RAID5system, there is little data write performancedeficiencies. In-depth analysis of the Linux kernel code and found the phenomenon thatdisk will have IO wait when dealing with write requests in the non-full stripe write, we useextra cache strategy to optimize the write performance. Handling a write request in thestrategy,will calculate the new parity information stored in the extra cache, to achieve theseparation of calculate the parity information and read or write data, eliminating disk I/Owaiting time to enhance the system performance. Tests show that the improved RAID5system, in the case of non-full stripe write has good performance optimization, solveproblem of the poor performance in little data write.