Research On The Critical Technology Of NAND Flash Storage Management Based On YAFFS2

With the development of storage technology, NAND Flash memory has been widely applied in various fields such as consumer electronics, automobile industry, aviation and aerospace for its hardware advantages, including high storage density, excellent I/O performance, low cost, low consumption, etc, and it may become the mainstream storage media of next generation instead of disk media. The development of Flash storage management technology plays an important role in making full use of flash hardware advantages and improving I/O performance of the system. As the extending of flash application, new scenarios have put forward higher demands on the storage management system, e.g. reliability, stability, and real-time performance. Based on the analysis of YAFFS2 storage management mechanism, this thesis focuses on the shortcomings of the typical flash file system YAFFS2 in three aspects, including I/O performance fluctuations caused by two different garbage collection patterns, the limitations of wear-leveling strategy, and the performance loss resulting from the synchronous metadata update scheme, and attempts to enhance the storage performance of YAFFS2 to sustain more complex applications. The main research work and academic achievements of this thesis are as follows:This thesis analyses the current situation of flash memory storage management technology, points out the shortages of FTL flash management system, and summarizes the advantages and characteristics by comparing three flash file systems. Besides, this thesis also analyses the storage management mechanism of YAFFS2 at the source code level, including the main data structure, file index structure, basic file operations, garbage collection algorithm, space allocation scheme, scanning process, then points out the shortages of garbage collection algorithm and wear-leveling strategy in YAFFS2.A pre-search based adaptive garbage collection algorithm is proposed. Aiming at the issue of sharply declined I/O performance caused by YAFFS2 two-level garbage collection patterns when the space utilization rate of flash is quite high, the algorithm adopts an overlapped segment searching method and dynamic valid-page threshold, which enhances the collection of dirtier blocks and concentrates the mainly work of garbage collection in passive pattern. Aiming at the decreasing of writing performance generated by wear-leveling operation, two different victim block selecting strategies are designed, and the algorithm adaptively selects one strategy to find the victim block for collection in accordance with the minimum average collection overhead from pre-search. The simulation results reveal that the proposed algorithm can effectively decrease the counts of aggressive garbage collection, improve the throughput of file system, and control flash wear evenness in some degree.For the local evenness issue of YAFFS2 wear-leveling strategy, this thesis designs a static wear-leveling algorithm based on block-erasing information. The algorithm firstly divides all blocks into two block groups, free block group and non-free block group. On the basis of the unevenness degree of current flash, the algorithm transfers the cold data to other clean pages and allocates blocks of the free block groups in a reasonable way to suppress the growth of maximum erasure counts. As the distribution of the block erasure counts in the physical space of flash is random, a probabilistic model is established to simplify the search operation in the block allocation process. Unevenness threshold is used as the trigger condition, which ensures the wear evenness degree is around a fixed value. The experimental results indicate that the algorithm strengthens the handling of static data, achieves global evenness in certain degree and improves the degree of wear evenness obviously compared with the original algorithm, which effectively prolongs the flash service life.Aiming at the excessive invalid flash metadata and low flash space utilization efficiency caused by YAFFS2 synchronous metadata update mechanism, a LIRS-based metadata cache scheme is proposed. According to the update information of metadata, different cache states are set up in the scheme, and the state transition mechanism is also proposed. Based on the influence degree of the metadata on the file system reliability, the cache replacement priority is divided into diverse levels. Moreover, a LIRS-based replacement algorithm is proposed in consideration of the asymmetric read/write cost characteristics of flash. As the feature of LIRS is retained, the replacement algorithm can capture the data with high accessing frequency. The algorithm preferentially replaces those caches that have a greater influence on the file system reliability or have lower I/O overhead. The experimental results show that the proposed scheme reduces the updating frequency of flash metadata and the direct writing operations on the flash, thus improves the management efficiency of metadata cache.