SRFTL: An Adaptive Superblock-Based Real-Time Flash Translation Layer for NAND Flash Memory

摘要

A NAND flash block must be erased before the I/O system rewrites the pages in it. When a full block (garbage) is erased, its valid pages should be copied to another place. Variable page copy operations cause uncertain garbage collection latency. Most of existing real-time flash translation layer (FTL) schemes use partial garbage collection to hide long write operation latency and provide a predictable worst-case response time (WCRT) upper bound. However, these schemes assign too much flash space to serve as write buffers for partial garbage collection even if the space will not be used in future. We propose an adaptive super block-based real-time FTL scheme (SRFTL) to reduce space waste, in which a fixed number of adjacent logical blocks are combined into a super block and are mapped to several physical blocks. The number of physical blocks corresponding to one super block is dynamically changed according to write requests on the super block. Furthermore, centralized and distributed partial garbage collection policies are adaptively used to reclaim victim blocks and reduce valid page copy operations. The SRFTL scheme is designed to improve average access time and reduce erase operations while guaranteeing the worst-case response time. Experimental results show that our solution can reduce 96.2% valid page copy operations, 47.9% erasing operations and 44.9% access time on average.