基于简单再生码的带宽感知的分布式存储节点修复优化

发布时间：2018-03-08 17:18

本文选题：分布式存储　切入点：简单再生码　出处：《软件学报》2017年08期 　论文类型：期刊论文

【摘要】：分布式存储系统为了保证可靠性,会采用一定的存储冗余策略,如多副本策略、纠删码策略.纠删码相对于副本具有存储开销小的优点,但节点修复网络开销大.针对修复网络开销优化,业界提出再生码和以简单再生码为代表的局部可修复码,显著降低了修复网络开销.然而,现有的基于编码的分布式容错存储方案大都假设节点处于星型逻辑网络结构中,忽略了实际的物理网络拓扑结构和带宽信息.为了实现拓扑感知的容错存储优化,相关研究在纠删码和再生码修复过程中结合网络链路带宽能力,建立树型修复路径,进一步提高了修复效率.但是,由于编码和修复过程的差异性,上述工作并不适合于简单再生码修复.针对该问题,结合实际物理网络拓扑结构,将链路带宽能力引入到简单再生码的修复过程中,对带宽感知的简单再生码修复优化技术开展研究.建立了带宽感知节点修复时延模型,提出了基于最优瓶颈路径和最优修复树的并行修复树构建算法,并通过实验对算法性能进行了评估.实验结果表明,与星型修复方式相比,该算法有效地降低了节点修复时延,提高了修复效率.
[Abstract]:In order to ensure the reliability of distributed storage system, some redundant storage strategies are adopted, such as multi-replica policy and erasure code strategy. Erasure code has the advantages of less storage overhead than replica. However, the network overhead of node repair is high. In order to optimize the cost of repairing network, the regenerative code and the local repairable code, represented by simple regenerative code, are proposed in the industry, which significantly reduces the cost of repairing the network. Most of the existing distributed fault-tolerant storage schemes based on coding assume that the nodes are in the star logic network structure, ignoring the topology structure and bandwidth information of the actual physical network. In the process of erasure code repair and regenerative code repair, the related research combines the network link bandwidth ability, establishes the tree repair path, and further improves the repair efficiency. However, because of the difference of coding and repair process, The above work is not suitable for the repair of simple regenerative codes. In view of this problem, the link bandwidth capability is introduced into the repair process of simple regenerative codes in combination with the actual physical network topology. In this paper, the simple regenerative code repair optimization technology for bandwidth awareness is studied. The bandwidth sensing node repair delay model is established, and the parallel repair tree construction algorithm based on the optimal bottleneck path and the optimal repair tree is proposed. The performance of the proposed algorithm is evaluated by experiments. The experimental results show that the proposed algorithm can effectively reduce the node repair delay and improve the repair efficiency compared with the star repair method.
【作者单位】：计算机软件新技术国家重点实验室(南京大学);上海市电力公司;
【基金】：国家自然科学基金(61373014) 国家电网科技项目(521104170019)~~
【分类号】：TP333

【相似文献】