HRSPC: a hybrid redundancy scheme via exploring computational locality to support fast recovery and high reliability in distributed storage systems

摘要

Replication and erasure codes are two popular schemes to provide fault tolerance in distributed storage systems. However, they both face some challenges when used in cloud storage. As the scale of data increases and failure becomes common in data centers, replication scheme suffers from high storage cost while erasure codes suffer from performance degradation and network bandwidth cost particularly under faulty conditions. In this paper, we present a novel scheme, named Hybrid Redundancy Scheme Plus Computing, or HRSPC for short, to achieve better performance and storage cost trade-off via exploiting the computational capability of storage nodes and taking advantage of both replication and erasure codes' features. It conducts degraded read or recovery with very low I/O cost and network bandwidth while the storage cost is relatively low, and thus to achieve high performance in both normal and faulty conditions. In addition, the reliability of HRSPC is higher than erasure codes and replication under similar configurations. (C) 2016 Elsevier Ltd. All rights reserved.