Clean-First Adaptive Buffer Replacement Algorithm for NAND Flash-Based Consumer Electronics

Abstract

NAND flash memory has some advantages including small size, high access performance, low power consumption, non-volatility, and shock resistance, so it has been extensively used as the dominant storage medium that serves as consumer electronics. Nevertheless, it has intense different physical hardware constraints with hard disks. Hence, buffer replacement algorithms used in hard disks has bad performance in NAND flash-based solid-state disk. In this article, we devise an excellent clean-first adaptive buffer replacement algorithm for NAND flash-based consumer electronics, which is named CF-ABR. It maintains four-page lists in LRU (Least Recently Used) order, which are a first used page list, a frequently used page list, and two replacement page lists. The first used page list holds the pages, which have been referenced once. The frequently used page list stores the pages, which have been referenced at least twice time. The other two replacement page lists are utilized to store the pages, which have been replaced from the buffer. To capture the frequency and recency of pages, CF-ABR assigns a variable called reference to each page in order to record its hit count and adjust the lengths of four-page lists dynamically. A battery of trace-driven experiments has been carried out in simulation environment and experimental results demonstrate that the devised CF-ABR algorithm performs better than existing buffer replacement algorithms containing the hit ratio, the number of read or write operations.