This project analyze the shortcomings of using IRM-based benchmarkings to evaluate cloud key-value stores. Also propose a trace replay model suitable for these systems, and develop KV-replay, an open-source replayer that implements this model.