Energy-Efficient Cache Management for NVM-Based IoT Systems
Yuanchao Xu, Yang Lu, Zeyi Hou, Qirun Huo, Keni Qiu · 2017
The advent of non-volatile memory (NVM) technologies drives the convergence of memory and storage towards a single-level store system, which can facilitate to reduce energy consumption. However, the assumptions made for previous well-established cache design might not hold anymore The previous well-established cache design, however, is inefficient in terms of energy since it assumes that memory is DRAM with symmetric read/write power as working memory other than persistent storage. To mitigate this issue, we propose two techniques, asymmetry-aware cache filling and locality-aware cache flushing. Using micro-benchmarks, the evaluation results show that our proposed policy can reduce power consumption by up to 24.90% compared with power-agnostic cache management policy.