Optimizing Repair-Cost of Locally Repairable Codes for Hot Data in Cluster Storage Systems

Abstract

Improving the repair performance of erasure code is a critical issue in order to maintain high data reliability in modern large-scale storage systems. Locally repairable codes (LRC) can improve the repair performance by locally repairing any single-node failure. In modern distributed cluster storage systems, the cross-cluster bandwidth is more scarce than the innercluster bandwidth. In this paper, we propose a well-designed placement strategy for LRC which is suitable for storing hot data. We show that our proposed placement strategy can reduce the cross-cluster bandwidth overhead in repairing node failures. Compared with the flat placement, the cross-cluster repair bandwidth can be reduced by more than 91.7% with our placement strategy in repairing one single-node failure, while the cross-cluster repair bandwidth reduction is 90.8% in repairing two-node failures.