TEA: A Traffic-efficient Erasure-coded Archival Scheme for In-memory Stores

To achieve good trade-off between access performance and memory efficiency, it is appropriate to adopt replication and erasure coding to keep popular and unpopular in-memory datasets, respectively. An issue of redundancy transition from replication to erasure coding (a.k.a., erasure-coded archival) should be addressed for unpopular in-memory datasets, since caching workloads exhibit long-tail distributions and most in-memory data are unpopular.

In this paper, we propose an encoding-oriented replica placement policy - ERP - by incorporating an interleaved declustering mechanism, and design a traffic-efficient erasure-coded archival schemes - TEA - for ERP-powered in-memory stores. With ERP in place, TEA embraces three salient features: (i) it alleviates cross-rack traffic raised by retrieving data-block replicas, (ii) it improves rack-level load balancing by distributing replicas via load-aware primary-rack-selection approach, and (iii) it mitigates block-relocation operations launched to sustain rack-level fault-tolerance. The empirical results show that TEA not only brings forth lower cross-rack traffic than four candidate encoding schemes, but also exhibits superb archival-throughput and rack-level-balancing performance. In particular, TEA accelerates archival throughput by at least 70.8%; and improves rack-level load-balancing by a factor of more than 1.58x relative to the four competitors.