FLRU:一种面向闪存固态盘的缓存管理算法

FLRU: A Caching Algorithm for Flash-based SSDs

本文针对固态盘的写代价高于读代价的特性,提出了一种面向闪存固态盘的缓存管理算法FLRU.其核心思想是为每一类数据页维护一个双缓冲区LRU队列:冷数据LRU队列,热数据LRU队列.提出的算法优先置换冷的干净数据页和冷的脏页(从冷数据LRU队列中置换),其次从热数据LRU队列中置换干净数据页,最后置换热的脏数据页LRU中的数据页.实验表明,与其它算法相比,FLRU算法能够适应不同类型的读写负载,同时明显降低了读写代价.

Existing algorithms adopted by operating systems consider the underlying devices as mechanical disks and ignore the asymmetry of read and write operations. And most existing flash-aware caching algorithms ignore the frequency of pages. In this paper,we propose a new caching algorithm for SSDs, called FLRU. It classifies pages into different types according to frequency and modification of pages. Each page type is organized by LRU. When selecting a victim page, FLRU replaces cold clean pages and cold dirty pages first and keeps balance between frequent pages and infrequent pages dynamically. We evaluate the performance of FLRU via tracedriven simulations. Experimental results show that FLRU can work well under different workloads, and reduce total read and write costs while avoiding significant degradation of hit ratio.