HeMatch: A redundancy layout placement scheme for erasure-coded storages in practical heterogeneous failure patterns

Abstract

Erasure codes are widely used in storage systems for providing data reliability due to the advantage of high storage efficiency, while the access efficiency becomes the main shortcoming because of the extra data retrieve and decoding for accessing unavailable data. Most existing work designs erasure codes based on the ideal failure pattern where all storage nodes fail at the same rate. However in practice, the physical storage nodes fail at different rates due to the heterogeneous hardware, topologies and application behaviors. In this paper, we consider the heterogeneous failure pattern and analyze how the failure pattern impacts the overall access efficiency and reliability of erasure-coded storage systems. We propose HeMatch, a redundancy layout placement scheme in practical heterogeneous failure pattern for erasure-coded storage access efficiency. Specifically, we first study how the heterogeneous failure pattern impacts the access efficiency and propose a general model based on the Tanner graph to evaluate and predict the access efficiency in specific failure pattern and redundancy layout. Then, we propose the redundancy layout placement scheme, which matches the redundancy layout with the physical storage nodes in practical heterogeneous failure patterns based on the evaluation and prediction from our model. The experimental results demonstrate that the model we propose accurately evaluates the access efficiency, and HeMatch saves unavailable data access cost by up to 20% and improves the system reliability as well.