A Study On Buffer Management And Wear Leveling Algorithm For NAND Solid State Drives

With the continuous advancement of large-scale integrated circuits and processor technologies,the difference in data processing speed between the storage system and the computing core significantly limits the overall performance of the computing system.The "storage wall" has become a key bottleneck in continuous improvement for system performance.The application of non-volatile memory technology represented by solid-state disk(SSD)provides a new idea for further improving the performance of the storage system.Although SSD have advantages such as high access speed and low power consumption,the disadvantages are particularly obvious,including asymmetric read and write,and a limited P/E times of flash memory blocks.This article focuses on the buffer management and wear leveling algorithms in NAND SSD design to improve SSD read and write performance and lifetime.For the problem of asymmetric read and write to flash,this paper design an Access-Characteristic Write Buffer Management(ACWB),which combines ISSP(Incremental Step Pulse Program)'s fast-reading,fast-write mode with write buffer,and uses novel access feature recognition methods to divide access requests into read-only,write-only,and interleaved types.For read/write-only pages,they will be directly processed in Flash,and operated with fast read/write mode to reduce the access delay.For interleaved read-write pages,they will be stored in the write buffer and hit directly in the buffer.The least write-proportional-replacement algorithm was used to eliminate dirty pages in the buffer area.The experimental results show that the ACWB and BPLRU,BPLRU+ buffer algorithms are compared from the three perspectives of access latency,erase count,and buffer hit rate,and this strategy achieves a very good performance improvement.For the problem of a limited P/E times of flash memory blocks,this paper design a Weighted Heap Sort Static Wear-Leveling Algorithm(WHWL).WHWL proposes a weight calculation method based on page data access frequency and block erase counts,which improves the hit rate of victim blocks and avoids unnecessary data migration operations.Secondly,a weight-based heap sorting victim block selection algorithm is proposed to speed up the screening of victim blocks.Experimental results show that compared with PWL and BET,using the same mapping mechanism,WHWL can increase the life span of solid state drives by 28%and 483%,and the degree of wear leveling and migration times are greatly improved.