Abstract:

In order to improve the life and overall performance of Solid State Disk(SSD),a novel page-level buffer management algorithm——CPBMS based on classification strategy is proposed.CPBMS divides Random Access Memory(RAM) into Hot Cached Data Page(H-CDP),Cold Cached Data Page(C-CDP) and Sequential Cached Data Page(S-CDP),which is responsible for buffering hot data page,cold data page and data page of requests with high spatial locality,respectively.In order to benefit from the high spatial locality of sequential requests,CPBMS prefetches multiple sequential data page into S-CDP,thus it can improve the response performance of sequential requests.When buffer is full,CPBMS preferred removes clean page of C-CDP.If C-CDP has no clean page,then replaces the dirty page,significantly reduce SSD read and write cost.Experimental results show that compared with page-level LRU and CFLRU algorithms,CPBMS algorithm significantly improves the buffer hit rate,effectively reduces the response time and flash block erasure times.

Key words: Solid State Disk(SSD), classification strategy, buffer management, Hot Cached Data Page(H-CDP), high spatial locality

摘要：

为提升固态硬盘的寿命和整体性能,提出一种基于分类策略的新型页级缓冲区管理算法。将数据页缓存分为热数据页存储区、冷数据页存储区和连续数据页存储区,分别缓存访问频繁的数据页、访问频率较低的数据页和连续请求的数据页。通过预取多个数据页到连续数据 页存储区,利用连续负载的高空间本地性提升对连续负载的响应性能。在缓冲区满时优先剔除冷数据页存储区中最近最少访问的干净数据页,若无干净数据页,再置换脏数据页,降低闪存的读写开销。实验结果表明,与页级LRU和CFLRU算法相比,该算法能明显提升缓冲区命中 率,有效减少响应时间和闪存块擦除次数。

关键词: 固态硬盘, 分类策略, 缓冲区管理, 热数据页缓存, 高空间本地性

CLC Number: 

• TP391