Abstract: Flash memory has been widely used in storage systems because of its nonvolatile nature, its small size, shock resistance and fast access speed. The traditional scheduler over flash memory storage systems is NOOP scheduler. There is much room for improving the I/O performance, especially over multi bank flash memory storage systems. Because several banks can operate simultaneously, we proposed a new scheduler called Multi-Bank flash memory Scheduler(MBS) based on the native file system YAFFS to take advantage of the parallelism of multiple banks by considering the high read speed. A flexible bank assignment policy was proposed to assign proper banks for write requests according to the attributes(hot or cold) of requests, which were identified by an AVI= based-tree mechanism. M13S scheduler reorders read and write requests and gives higher priority to reads. The experimental results show that the I/O throughputs and average response time are improved significantly compared with the NOOP scheduler. An even erasable count and capacity utilization were obtained between different banks in the multi-bank storage system.

Key words: Multi-bank flash memory, I/O scheduler, Native flash memory file system, Dynamic bank assignment policy, Identification of request attribute