摘要

An efficient page replacement algorithm, which is named EPRA and customized for different kinds of NAND flash memories with different cost ratios of write operation to read operation, is proposed in this paper. Each dirty victim page candidate within the main memory is divided into a fixed number of flash pages. EPRA assigns a weighted value to each victim page candidate and selects a victim page candidate that has the least weighted value as a victim page. The elapsed time of each dirty flash page since the most recent reference is used to classify the dirty flash pages within the dirty victim page into the hot dirty flash pages and cold ones. When a dirty victim page is selected, EPRA only writes the dirty flash pages within the victim page back to NAND flash memory by separating the hot dirty flash pages from the cold ones and writing them to different free blocks within the NAND flash memory. Trace-driven simulations show that the proposed EPRA algorithm is superior to existing page replacement algorithms when they are performed on different kinds of NAND flash memories(1).