非均匀故障保护的分组修复码构造

考虑到实际分布式存储系统中存在热度不同的文件，构造了一种基于非均匀故障保护的分组修复码（GRC-NFP），可对热文件和高故障概率节点提供更高等级保护，并降低多故障节点修复的磁盘读取开销.在文件冷热分组后，用所存目标节点故障概率表征数据块故障概率，并排序，存入长度依次递增的多个数据分组，并生成组编码块.性能分析和实际系统部署结果表明，与里德-所罗门码和分组修复码相比，GRC-NFP可在存储开销较小的条件下拥有较高的容错能力和较低的修复局部性，并且使热文件能够受到更有效地保护.系统部署下较少的编码和故障修复时间进一步证明了GRC-NFP的可行性.

Construction of Group Repairable Codes for Non-Uniform Fault Protection

Considering that there are files with different heat in actual distributed storage systems, a class of group repairable codes based on non-uniform fault protection (GRC-NFP) is proposed. GRC-NFP provides higher protection for hot files and nodes with high fault probability, and reduces the disk I/O overhead for repairing multiple failed nodes. Specifically, after hot and cold grouping, the fault probabilities of data blocks are represented and sorted by that of the stored target nodes. Data blocks are stored into multiple data groups with increasing lengths, and group encoded blocks are further generated. Performance analysis and actual system deployment showed that GRC-NFP had higher fault tolerance and lower repair locality under less storage overhead compared with Reed-Solomon codes and group repairable codes. Moreover, the hot files can be protected more effectively by adopting GRC-NFP. The fewer coding and fault repair time under system deployment further proved the feasibility of GRC-NFP.