當資料具有隨機存取的性質時,如果我們使用傳統硬碟作為存儲裝置,那麼當需要的資料與目前讀寫頭的位置有很大距離,將會耗費很多時間。然而,在固態硬碟上卻不會有這種現象,所以一般具有隨機存取性質的資料會存在固態硬碟中,而且固態硬碟具有非揮發性、低耗電、抗震性、存取快速等優點。 雖然固態硬碟中 sequential access 和random access的效能每年都有提升,但是write/read操作的sequential 和random改善的比值卻受到限制。本論文將著重於FTL(flash translation layer)此階層對random I/O效能的改善。我們將FTL演算法分為direct mapping和log-based mapping以及demand-based mapping三種類別。並且我們藉由分析random access的效能,說明不同演算法的利弊得失。
The performance of sequential access and random access of SSDs (solid-state disks) is improved every year, but the improved ratios of read/write operations are limited. The focus of this thesis is to survey how a flash translation layer (FTL) improves the performance of random I/O. We group various algorithms of FTL into three categories named direct mapping, log-based mapping and demand-based mapping schemes. An analysis of random access performance is given to illustrate the benefits and costs of different algorithms.