Research On The Repair Of Failure Node In Distributed Storage Systems

With the arrival of the massive data era, distributed storage technology has become an important method to solve the problem of massive data storage, with its high scalability and low cost. Due to the large scale of the distributed storage systems, the problem of node failure caused by the failure of the storage disks or other network firmware has become a common situation. In order to ensure the availability of data, generally, it is necessary that a part of redundant data to be stored in the nodes. Beside, in order to maintain the high reliability of distributed storage systems, we need to repair the failure storage node in time. On the other hand, the erasure codes is usually used to improve the reliability of the distributed storage systems, and compared with the strategy of replication, erasure codes can achieve the best storage expenses in distributed storage systems, but it will restore the entire original file when repairing a single failure node. While ensure that a system has low storage costs, Regenerating Codes can be better to reduce the bandwidth overhead caused by the failure of repairing, which has important application prospects. Meanwhile, it is also very important to think about how to reduce the disk reading overhead caused by the failure of repairing. In order to reduce the repair bandwidth overhead and disk reading overhead when repairing the single failure node of a distributed storage system, this paper mainly studies the following contents:(1)While the Simple Regenerating Codes has been studied and improved, a new segmentation coding scheme is proposed in this paper. Concretely, the f coded blocks with the same index are divided into two sections, and then XOR the coded blocks of each section to generate a new parity block. Compared with Simple Regenerating Codes, the proposed segmentation coding scheme based on Simple Regenerating Codes can optimize the repair bandwidth and disk reading overhead, while at a little expense of storage overhead.(2)The Locally Minimum Storage Regenerating Codes is proposed. On the basis of the algorithm of construct code block gradually based on Minimum Storage Regenerating Codes, firstly the nodes of the distributed storage system are divided into groups, and next, for each sub file, a smaller size of the Minimum Storage Regenerating Codes is constructed in each group. Thus, for the single failure node, it can be locally repaired by the algorithm of construction code block in each group, moreover, the grouping scheme of Locally Minimum Storage Regenerating Codes under different conditions has been given in this paper. Compared with the algorithm of construct code block gradually based on Minimum Storage Regenerating Codes, Locally Minimum Storage Regenerating Codes can reach the minimum storage overhead, and it can reduce the repair bandwidth and disk reading overhead effectively.