基于树结构的纠删码数据更新策略  

Erasure-coded data update strategy based on tree-structure

作　　者：赵银瑞 殷双飞 陕振[1] 王梓斌[2] ZHAO Yin-rui;YIN Shuang-fei;SHAN Zhen;WANG Zi-bin(Institute 706,Second Research Academy of China Aerospace Science and Industry Corporation,Beijing 100854,China;National Key Laboratory of Science and Technology of Information System Security,Beijing Institute of System and Engineering,Beijing 100101,China)

机构地区：[1]中国航天科工集团第二研究院706所,北京100854 [2]北京系统工程研究所信息系统安全技术重点实验室,北京100101

出　　处：《计算机工程与设计》2023年第3期725-731,共7页Computer Engineering and Design

摘　　要：为保证存储集群中数据的可靠性,缓解跨机架环境中单节点的有限可用带宽,提出一种纠删码的数据更新策略DBTU(data delta log based tree-structure update)。该方法包含2个设计要点:数据块更新采用就地更新策略,校验块更新采用基于数据增量日志的推测性部分写入策略,减少读后写操作;采用自顶向下的树结构传输更新数据,缓解单节点的带宽瓶颈。实验结果表明,在100 Mbps带宽下,DBTU的更新吞吐率相较FO(full overwrite)和PARIX(speculative partial write)分别提高了54.3%和51.6%。To ensure the reliability of the data in the storage cluster and alleviate the limited available bandwidth of a single node in a cross-rack environment,a data update strategy of erasure code,namely DBTU(data delta log based tree-structure update),was proposed.DBTU included three design points.The data block update adopted the in-place update strategy while the parity block update adopted speculative partial write based on data-delta log to reduce write-after-read operations.Tree-structure with top-down transmission was adopted to relieve the bandwidth bottleneck of a single node.Experimental results show that under 100 Mbps bandwidth,the update throughput of DBTU is 54.3%and 51.6%higher than that of FO(full overwrite)and PARIX(speculative partial write),respectively.

关 键 词：纠删码 存储集群 更新树 数据更新 增量 跨机架 奇偶校验 

分 类 号：TP302.8[自动化与计算机技术—计算机系统结构]