SlimFTL: a Small and Fast Page-level FTL using Hash Functions

Abstract

As the capacity of an SSD increases, the amount of DRAM for managing the SSD increases proportionally. Since the DRAM cost directly affects the overall SSD price, it is important to minimize the DRAM size without degrading the SSD performance. In this paper, we propose a novel hash-based FTL mapping technique, SlimFTL, that meets this goal. SlimFTL overcomes the GC inefficiency problem of an existing hash-based FTL in two directions. By employing an efficient indirection layer between the logical page and its hashed physical block, SlimFTL reduces the block copy overhead during GC. SlimFTL exploits the spatial sequentiality among successive writes so that sequential writes can be mapped to the same physical block, which significantly reduces the number of valid copies during GC. Experimental results show that SlimFTL can achieve the same performance level of a page-level mapping scheme with only 44% of the DRAM capacity.