Abstract

In recent years, erasure codes have become the defacto standard for data protection of large scale distributed cloudstorage systems at the cost of an affordable storage overhead. While traditional erasure coding schemes, such as Reed-Solomoncodes, suffer from high reconstruction cost and I/Os. The recentpast has seen a plethora of efforts to optimize the tradeoffbetween the reconstruction cost, I/Os and storage overhead. Quietly different from all prior studies, in this paper, our erasurecoding technology makes the first attempt to take advantage ofthe unequal failure rates across the disks/nodes to optimize thereconstruction performance and system reliability. Specifically, our proposed technology, the Unequal Failure Protection basedLocal Reconstruction Code (UFP-LRC) divides the data blocksinto several unequal-sized groups with local parities, assigningthe data blocks stored on more failure-prone disks/nodes into thesmaller-sized group, so as to provide unequal failure protectionfor each group. In this way, by exploiting the nonuniform localparity degrees, the proposed UFP-LRC enables the data blocksthat are stored on more failure-prone disks/nodes to tolerate agreater number of failures while suffer from less repair costthan others, leading to a substantial improvement of overallrepair performance and reliability for cloud storage system. Weperform numerical analysis and build a prototype storage systemto verify our approach. The analytical results show that the UFPLRCtechnique gradually outperforms LRC along the increaseof failure rate ratio. Also, extensive experiments show that, whencompared to LRC, UFP-LRC is able to achieve a 10% to 13%improvement in throughput, and a 8% to 12% reduction indecoding latency, while retaining a comparable overall reliability.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call