DSSR:一种纠删码中用于数据重构的数据源选择算法
首发时间:2018-12-07
摘要:分布式存储系统中,纠删码是一种用低冗余来保证数据可靠性的非常有效的方法。但是,纠删码在重构数据时主要通过获取同一条带中的k个块重构故障节点上的不可用数据。另外,分布式存储系统的网络拓扑通常比较复杂,因此不同节点间的通信延迟是不同的。然而,当实际存储系统在选择数据源来修复不可用数据时,它们通常没有考虑网络情况而是随机选择同一条带中k个存活的节点来传输数据。由于修复时间是由最慢的节点决定的,因此随机选择数据源的过程可能会选择性能较差的节点从而影响整个修复过程。因此,为了减少数据重构时间,我们提出了DSSR,一种基于网络情况选择数据源的算法。DSSR是通过选择具有最小通信延迟的前k个节点来重构不可用数据。实验结果显示相比与随机选择k个存活节点修复数据的方法,DSSR能有效地减少数据修复时间。
关键词: 分布式存储 纠删码 单节点故障 网络延迟 重构 数据源
For information in English, please click here
DSSR: An Algorithm for Data Source Selection to Reconstruct Data in Erasure Codes
Abstract:Erasure codes are an effective means to maintain data availability with low redundancy for distributed storage systems. However, erasure codes mainly work by taking any k (referred to the number of data chunks) surviving nodes of a stripe to recover unavailable data on the failed node. In addition, because the network topology of distributed storage system is often complex, the communication delays between different nodes are different. However,when selecting data sources to recover unavailable data in the practical storage systems, it is usually random to select k surviving nodes to transform data without considering the network. Because the recovery time will be determined by the slowest node, the process of randomly selecting data sources may select nodeDSSR: An Algorithm for Data Source Selection to Reconstruct Data in Erasure Codess with poor performance to affect the whole data repair process. To reduce the reconstruction time, we propose DSSR, an algorithm for choosing data sources to reconstruct data based on the network. DSSR chooses the first k nodes having minimal communication delay to reconstruct the lost chunks at single-stripe level. The result shows DSSR can effectively reduce the reconstruction time comparing with the method of randomly choosing k surviving nodes of a stripe.
Keywords: Distributed storage Erasure codes Single failure Network delay Reconstruction Data Source
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
DSSR:一种纠删码中用于数据重构的数据源选择算法
评论
全部评论0/1000