Wear-Leveling Aware Page Management for Non-Volatile Main Memory on Embedded Systems

摘要

Non-volatile Memories (NVMs), have many promising characteristics, such as low leakage power, low cost, non-volatility, and high scalability, which are all attractive for embedded systems to employ them as the main memory. However, one of the constraints that undermine the credential of NVMs as main memory is its limited write endurance. To tackle this problem, this paper proposes five techniques: Rearrangement Inequality Based Page Allocation (RIPA), Virtual Page Mapping (VPM), On-demand Memory Merging and Splitting (OMS), Periodical Page Swapping (PPS), and Normalized Boundary Calibration (NBC) to evenly distribute the writes on Nonvolatile Main Memory (NVMM) purely on the Operating System (OS) level, which can greatly extend lifetime of NVMM. Without extra hardware support, OS management is easy to be integrated into existing embedded systems. The experimental results show that with less then 0.6 percent performance overhead the proposed techniques can extend the lifetime of NVMM to 17.28 times longer compared with traditional methods.