Quail: Using NVM write monitor to enable transparent wear-leveling

Abstract

Abstract Emerging non-volatile memory (NVM) technologies promise persistence, byte-addressability, low power consumption, and high density. However, NVM suffers from limited write endurance; skewed writes can extremely curtail the lifetime of NVM. Although a number of wear-aware allocators and wear-leveling algorithms have been proposed to address the wear-out problem, they come with limitations in the accuracy of NVM write count or compatibility with existing hardware and software. In this paper, we implement four software-based techniques of monitoring NVM writes and obtaining the write counts, which are more accurate and non-intrusive for existing programs. In particular, we propose a novel adaptive monitoring technique to alleviate the monitoring overhead while maintaining the count accuracy. Based on our monitoring tool, we further provide a swap-enabled transparent wear-leveling mechanism in order to even out the NVM writes. We conduct extensive experiments to verify the performance of our proposed NVM monitor and wear-leveling mechanism. The results show that the monitoring methods we implement can obtain high accuracy at an affordable performance overhead; our proposed wear-leveling mechanism can reduce the maximum NVM write by 89% when compared to its no wear-leveling counterpart, with less than 4% time overhead.