Reducing Write Operations in Flash Memory Using a Modified CFLRU Page Replacement Algorithm

Abstract

Modern computing system heavily depends on the flash-based memory system. Effective cache management is fundamental to improving performance and extending lifespan of flash memory. Traditional page-replacement algorithms, such as LRU, were originally designed for magnetic disks and do not fully deal with the unique characteristics of flash memory, including asymmetric read and write speeds and limited erase operations. This article introduces an enhancement to the Clean First LRU (CFLRU) algorithm, termed Dirty Hot Clean Hot CFLRU (DCH-CFLRU), which provides both clean and dirty pages with a second chance. The proposed solution aims to reduce write counts while upholding a stable hit ratio. Experimental outcomes show that DCH-CFLRU reduces write operation by up to 14% compared to CFLRU through multiple datasets specially in write most cases. The modified algorithm exhibits better overall flash memory performance and extending its lifespan.