一种三容错数据布局  

A Data Placement Based on Toleration Triple Failures

作　　者：万武南[1,2] 王拓[1] 索望[1] 

机构地区：[1]成都信息工程学院网络工程学院,成都610225 [2]电子科技大学计算机科学与技术学院,成都610054

出　　处：《电子与信息学报》2013年第10期2341-2346,共6页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(60873216);四川省教育厅重点项目(12ZA223)资助课题

摘　　要：随着存储介质的增多,单容错、双容错的数据布局方案已经无法满足现有分布式存储系统对可靠性要求。该文在双容错行对角奇偶校验(Row Diagonal Parity,RDP)码的基础上,提出一种新的扩展行对角奇偶校验(Extending Row Diagonal Parity,E-RDP)码,能够容许任何3存储节点出错,具有最大距离可分(Maximum Distance Separable,MDS)编码特性,冗余率与纠错能力达到3容错编码最优。并采用不同斜率几何直线图描述编译码过程,给出了一种快速译码算法,易于软硬件实现。与其它纠删码数据布局方案进行比较,理论分析结果表明,E-RDP码的空间利用率、编译码效率、小写性能以及平衡性的综合性能达到最优,具有实用价值。With increase of storage devices, the data placements based on toleration single or double failures can not meet the requirement of the reliability in the distributed storage systems. On the basis of the Row Diagonal Parity （RDP） code for double toleration failures, a new class of array codes for triple storage failures is presented which is called Extending Row Diagonal Parity （E-RDP） code. The E-RDP code has the Maximum Distance Separable （MDS） property, and it is optimal in redundancy rate and erasure correcting capability among triple erasure-correcting codes. The procedures of encoding and decoding are depicted by geometrical lines of different slope, then a fast decoding algorithm is given and it is more easily implemented by software and hardware. The theoretical analysis shows that the comprehensive properties of the E-RDP code such as encoding and decoding efficiency, small writes and balance performance, are better than other popular MDS codes, thus the E-RDP code is practically meaningful for storage systems.

关 键 词：数据存储 编码 纠删码 行对角奇偶校验（RDP）码 可靠性 

分 类 号：TP333[自动化与计算机技术—计算机系统结构]