RDP 码存储系统节点故障恢复的读盘优化

摘要

实际的分布式存储系统面临着频繁的磁盘故障。为了保障数据可靠性，纠删码被广泛地部署在大规模存储系统中。在基于纠删码的存储系统中，快速有效地修复故障磁盘上的数据对于维护数据可靠性有重要意义。研究最重要的容两错纠删码———RDP（Row-diagonal parity）编码的磁盘故障修复问题，优化修复过程中磁盘访问的连续性。提出的单磁盘故障修复方案在保证读取数据量最小的前提下，最大程度避免了磁盘数据的随机读取，保持数据读取的连续性。通过在实际的分布式存储系统中实验，验证了该修复方案的实际性能，证实该算法可以很好地改善混合修复方案的随机读取引起的修复速度下降问题，最终提高了修复效率。%Real distributed storage systems face frequent disk failures.In order to ensure data reliability,erasure codes are widely deployed to large-scale storage systems.In an erasure coded-based storage system,to quickly and efficiently recover the lost data in failed disk has great significance to maintaining the data reliability.In this paper,we study the disk failure recovery problem for storage systems with RDP,one of the most important double-fault tolerant erasure codes,and seek to optimise the disk access sequentiality during the failure recovery.The recovery scheme for single disk failure we proposed avoids to the greatest extent the random reads of disk data on the premise of ensuring the minimum amount of data being read,and also preserve the sequentiality of data reads.Through the experiments in a real distributed storage system we validate the practical performance of the proposed recovery scheme,it approves that our recovery algorithm can well meliorate the problem of recovery speed slowing down in hybrid recovery scheme caused by random disk accesses and eventually improves recovery efficiency.