网格环境下基于编码机制的数据复制研究

Research on the Coding Mechanism-based Data Replication in Grid Environments

本文提出了基于编码机制的网格数据复制思想,通过对副本数据进行线性分组编码,并将其分散保存到网格存储节点,可形成具有纠删能力的编码子副本组。针对目前热点研究的线性分组编码,探讨基于Cauchy Reed-Solo-mon Code、Tornado Code和Random Linear Code的编码数据复制方案,通过建模手段讨论三者的副本数据访问性能和副本数据可靠性,并与传统的完整数据复制和分块数据复制进行对比分析,证明所提出的编码数据复制有着较优的综合性能。具体实验数据进一步说明,编码副本的编码开销占整个数据复制开销的较小比例,表明编码数据复制是具有可行性的技术方案。

This paper proposes a thought of coding mechanism-hased grid data replication. By carring on the linear block coding to the replica data and dispersedly saving them to the grid storage nodes, erasure capacity can be formed with the coded child-replica group. Focusing on the current hot linear block coding, this paper presents the schemes of the coding data replication based on the Cauchy Reed-Solomon code, Tornado Code and Random Linear Code, and discusses the data access and reliability performance of the replicas by the modelling method. Compared with the complete data replication and block data replication, the coding data replication has the overall better performance. Specific ex- perimental data further illustrate that the coding expenses of the coded child-replica account for a relatively small proportion of the data replication, and indicate that the coding data replication is a feasible technical scheme.