Reducing access latency in erasure coded cloud storage with local block migration

摘要: Erasure coding has been applied in many cloud storage systems to enhance reliability at a lower cost than replication. While large amount of prior work aims recovery performance and reliability, the overall access delay coded still needs be optimized. As most production adopt systematic code place original copy each block on only one server read normally, it is harder balance loads more likely incur latency tails three-way replication, where can from any 3 servers storing block. In this paper, we propose reduce by moving blocks with anti-correlated demands onto same for statistical load balancing. We formulate optimal placement as problem similar Min-k-Partition, local migration scheme, derive an approximation ratio function demand variation across blocks. Based request traces Windows Azure Storage, demonstrate that our scheme significantly few moves, especially when skewed.