FastDR: Boosting Degraded Reads in Heterogeneous Erasure-Coded Storage Systems

Introduction

To ensure data availability, erasure codes have been widely deployed in large-scale storage systems. In reality, temporary failures, which result in degraded reads, contribute to the majority of node failures. Thus, degraded reads become performance critical operations in large-scale storage systems. On the other hand, due to system upgrade, storage nodes tend to be heterogeneous with different storage capacities and I/O bandwidths. To address the challenge, we propose FastDR, a scheme that addresses node heterogeneity and exploits I/O parallelism, so as to boost the degraded read performance in the presence of temporary failures. FastDR uses a heuristic approach to search for the efficient degraded read solution timely.

We implement a FastDR prototype and further evaluate it quantitatively atop a Hadoop distributed file system cluster with 10 storage nodes. We demonstrate that our FastDR achieves efficient degraded reads when compared to the basic degraded read approach for erasure-coded data.

Publication