关键词: 固态硬盘（SSD）, 缓存, SSD寿命, 访问序列折叠, 缓存替换

Abstract: Abstract: Because of the limited write endurance of solid state drives (SSDs), the write amount of SSD cache becomes another important critical metric to measure cache replacement algorithms except for the cache hit rates. Therefore, how to improve the overall quality of the cached data for longer lifetime of SSDs, i.e., promoting the efficiency of transforming cache writes into cache hits, is very important for the cache replacement algorithms. At present, most of the existing cache replacement algorithms reply on temporal locality, i.e., the recently accessed data usually have a high possibility to be requested soon with the result of requiring frequent data updates and a high write pressure for SSDs to ensure a high hit rate. Some improved algorithms prevent some least accessed data reducing write amounts through a high cost. A solution aiming at improving the overall quality of cached data based on the observed long-term law of data popularity with low overhead is required. This paper proposes a cache replacement algorithm called folded access sequence (FAS) for the SSD read cache. FAS is designed to identify the long-term hot data with only a low overhead, leading to higher quality of cached data in SSDs, high hit rates, reduced amounts of written data to SSDs, and longer SSD lifetime. The experimental results show that FAS can reduce the SSD writes by 90% compared with the traditional LRU (least recently used) algorithm and the hit rate loss does not exceed 10%. Compared with the improved cache algorithms like SieveStore and L2ARC (level 2 adjustable replacement cache), FAS's write amount is reduced for more than 50% with similar hit rates. The results exhibit that the proposed FAS can effectively keep high-quality cached data, reduce the written amounts to SSDs, and extend the SSD lifetime.

Key words: solid state drive (SSD), cache, SSD lifetime, folded access sequence, cache replacement