Analysis And Optimization Of The UBIFS File System For Flash Memory

As an external storage medium frequently utilized in the embedded system, Flash Memory has several advantages such as non-volatility, resistance to shock, low power consumption, and small size, etc. It mainly includes two types:NAND Flash and NOR Flash, in which the former is more widely adopted in the daily consumer electronics, because of its rapid growth in the capacity and the progressive promotion in the performance-price ratio.The widespread usage of flash memory is not only determined by its merits and disadvantages of physical properties, but lies in the quality of data management in the file system for flash memory. Recently, the traditional system has been far away from the customers’ demand in large-scale NAND flash environment, for instance, the time of mounting, the RAM consumption, the I/O performance, the wear-leveling, and other highly-expected index. As the new generation file system for flash memory, UBIFS is a log-constructed one on ground of Linux operating system which solves the technological bottleneck problems in the utilization of NAND Flash as a Memory Technology Device, possessing extensive application prospect. Concentrating on the analysis and research of UBIFS, the thesis designs two optimized schemes at quick mounting and hot data identification aspects, through which the test proves that the promoted UBIFS achieves the expected target, displaying extensively its academic significance.Given the inadequate Chinese literature targeted to UBIFS, this study attempts to analyze the UBIFS file system. It introduces the systematic structure of UBIFS, pointing out that UBIFS is comprised in the modules of UBI and UBIFS from whose perspective the working principle is elaborated. Meanwhile, this thesis explores meticulously the key technologies in mapping management, wear-leveling and file index.Surrounding the problems in the analytical procession, the thesis proposes the optimization schemes from two levels. The first lies in the UBI modular level, indicating the strategy of enhancing the start-up speed on the sacrifice of reducing the speed of shut-down. In order to establish the block association table, mounting UBI requires scanning the header information of all psychical blocks. Regretfully, with the enlargement in the capacity of flash memory, the time for scanning the whole storage increases with the consequence of slow-down in starting the embedded system, which confines the application prospect of UBIFS in the field of large-capacity flash memory. If the header of all the psychical blocks could be integrated and reserved in the obligate block when uninstalling the UBI module, the header can be getted from the reserved blocks with next mounting UBI module, so as to accelerate the speed of starting-up. This optimization is testified to prove that it could enhance profoundly the mounting speed of UBI module.In terms of the UBIFS module, the written data is divided into three catalogues, to which the distinctive allocation principles of physical modules will correspond respectively. However, the methodology of data classification is static rather than dynamic, so that it could not identify the hot data in the user data and further influences the performance of wear-leveling. Through reference to the main memroy page-replacement arithmetic, the thesis proposes two-level LRU (Least Recently Used) algorithm which enables the UBIFS to identify the hot data dynamically. The tests show that the two-level LRU algorithm could improve the performance of wear-leveling in the UBIFS file system.