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 t