Research And Implementation Of Reconstruction Optimization For RAID Based On I/O Workload Migration

With the impetus of information technology, unrecoverable catastrophic failures arenot allowed in important storage systems, such as the storage systems in military, financial,and large enterprise. Therefore, reliability is an important factor to measure the overallperformance of storage system. RAID (Redundant Array of Independent Disks) canimprove the reliability of storage system. However, if another disk failure occurs at theprocess of reconstruction, except RAID6, all the data in this storage system will be lost.The longer the time of reconstruction, the higher probability of data loss is. Therefore,improving the speed of reconstruction is an important measure to improve the reliability ofRAID, and how to improve the speed of reconstruction effectively is an important topic.The on-line reconstruction of RAID is closely related to I/O workload due tocontention for the shared disk bandwidth. I/O workload can influence the performance ofon-line reconstruction of RAID seriously. Therefore, an optimization of reconstructionbased on workload migration has been designed and implemented. According to theprinciple of locality, we can find the hot spot of read request in block level. The readrequests of hot spot and write requests are redirected to proxy storage device at thereconstruction period. The redirected write data will be migrated back after reconstructionto ensure the correctness of RAID. This strategy can reduce the I/O workload of RAID,which leave more bandwidth to reconstruction and improve the speed of reconstruction.What’s more, most of I/O workloads are sequentially serviced by proxy storage devicethat workload is much slight, reducing the average response time significantly.Optimize the PR algorithm based on workload migration in RAID5and test theperformance of reconstruction using different workload. The evaluation shows that thereconstruction time and average response time is improved significantly under theoptimized algorithm, especially for the workload which has intensive write request.