Abstract: NAND Flash memory can’t be managed by traditional file systems for its special access characteristics.Although the problem is solved by using FTL in NAND Flash devices,the structure and algorithm in traditional file systems designed for disk can easily cause severe performance degradation and uneven wear in NAND Flash devices.Flash file systems that combine some FTL functions can make better use of NAND Flash memory with further optimization.UBIFS is a widely used Flash file system in Linux,but there are still severe write-amplification and frequent garbage collection in the storage system using UBIFS.In order to solve these problems of UBIFS,this research used hash table with multi-hash functions to identify hot data in UBIFS to reduce identification overhead and improve recognition accuracy.By using multiple log,cold and hot data are separated into different storage area to reduce the frequency of garbage collection.The delay commit technology of log was also adopted to reduce the number of times write-amplification occurs caused by frequent log commit.A series of experiments were conducted to verify the performance of the proposed method.Experiments show that internal writes to physical block in optimized UBIFS are reduced by 5%~10% compared to that in the original UBIFS,and the times of garbage collection triggered in optimized UBIFS were reduced by 7%~13% compared to that in the original UBIFS.In general,IOPS of storage system is improved by 5%~18% and system performance degradation is effectively alleviated by using optimized UBIFS.

Key words: NAND flash,UBIFS,Hot data,Garbage collection