JUMPRUN: A hybrid mechanism to accelerate item scanning for in-memory databases
Hong-Yeol Lim, Gi-Ho Park · 2017
In this paper, we introduce a simple and effective mechanism, called JUMPRUN (jump and run), to accelerate the scanning operation of KEY-VALUE datasets and alleviate the massive data traffic across the memory hierarchy. The proposed JUMPRUN incorporates a software acceleration scheme as well as a dedicated hardware accelerator designed on the near-memory processing (NMP) concept. In evaluation results, compared to a conventional Memcached system with stacked memory layers, the proposed JUMPRUN mechanism provides an average 8.4 times higher performance in the scanning operation and also delivers average 1.37 times overall speedup. The proposed mechanism also delivers a significant reduction of data traffic in both the on-chip memory hierarchy (72%) and stacked DRAM layers (59%), compared to a conventional Memcached system.