Research On NAND File System

NAND flash is widely used in mobile phones, portable music players, digital cameras, and more embedded system, provide these devices with low cost and reliable storage capabilities. NAND flash technology is driven mostly by desire of increased capacity and performance. The new type MLC NAND provide high density, so chip cost lower than the old type SLC NAND.NAND suffers from two limitations: bits can only be cleared by erasing block, each block endure a limited number of erase cycles, usually between 10000 and 100000,beyond that the block cannot store data reliably; due to the reliability issues, ECC is used to identify and correct errors during the operations. For SLC NAND, the most uesd ECC algorithm can detect and correct single bit errors, however MLC error rates increase, need to perform multiple bit correction.The sophisticated data structures and algorithms are needed to use NAND stroage system. There are two approaches to address NAND specific restrictions: one is to use traditional disk file system and Flash Translation Layer, like FAT+NFTL; second more effective approach is to use a flash file system on the raw flash, like JFFS2, YAFFS.This paper research the special designed for NAND file system YAFFS, from the view point of data structure, mount speed, wear-levelling and garbage collection. Porting YAFFS to demo board, evaluate YAFFS2 and its old version YAFFS1 performance.As the NAND capacity is increasing rapidly, flash file system mount speed is more important than ever. There is no fixed area to store file system index info, metadata is distributed to NAND page, so the mount process of YAFFS1 need to scan all flash page, suffers from long time, a snapshot methord is used in YAFFS2, short the time much, but after the power fail this methord is useless, also need to scan, and file system maybe in a inconsistent state.A log record methord is used to slove this problem in this paper, and a methord is designed to detect power fail. The log record is used to trace the file system changement, combine with the snapshot methord, after power fail only need to scan the flash page which store log record, considering the wear-levelling problem the log record is not stored in fixed area. The experiment shown that, this methord mount time is faster than old scan methord, and keep the file system reliable.