Efficient Updates in Erasure-Coded Storage Systems

Vertical array codes are widely used in erasure-coded storage systems due to their optimized update complexity and prominent coding performance. However, in actual data center, the update performance of the storage system is often limited by the cross-rack bandwidth. We propose a new update algorithm that aims to mitigate cross-rack traffic caused by data updates in distributed storage based on vertical array codes. Our proposed algorithm mainly contains three parts: (i) node grouping, which matches and combines the data nodes by comparing the associations between the nodes; (ii) intra-rack aggregation, which reduces the traffic by intra-rack aggregation of the relevant data blocks; (iii) intra-rack update, which converts a portion of cross-rack transfer into an intra-rack transmission. We conduct experiments and present the performance comparison with multiple vertical array codes. Our experimental result shows that it can significantly reduce cross-rack traffic generated by data updates and enhance system throughput.