Dynamically Reconfigurable Architecture for High-Throughput Hash Function in Key-Value Store

Distributed in-memory key-value stores (KVSs), such as Redis and Memcached, are widely deployed in modern data centers. In Redis, the throughput of Murmurhash2 is important to the performance of the whole KVS, and further affects the whole server. In this paper, we propose a high-performance scalable architecture for Murmurhash2 to improve data locality and reduce the latency, and deploy partial reconfiguration scheme to maximize the ratio of performance/area. Experiment results on Xilinx SOC platform demonstrate that the proposed architecture can improve the throughput by approximate 1.4x and achieve up to 5.3x performance/watt gain compared with the typical Intel platform.