Research On FTL Algorithm For Optimizing Read Performance Of High-capacity SSDs

With the rapid development of flash memory technologies such as 3D stacking and multi-bit cells,the capacity of widely used flash-based SSDs is constantly increasing and the cost is constantly reducing.The address translation of the SSD is conducted by the Flash Translation Layer(FTL)and a mapping table is maintained to translates the logical addresses of the host to the physical addresses of the flash.The size of the mapping table increases with SSD capacity increases,but the growth of SSD's built-in DRAM capacity cannot keep up with that of the SSD capacity.Thus,it's not feasible to cache the entire mapping table in the built-in DRAM to accelerate the address translation.The traditional FTL algorithm stores the complete mapping table on flash memory and caches only a fraction of the mapping table.A cache miss causes extra flash memory access overhead.Its mapping cache hit rate depends on the locality of the workload.The traditional FTL algorithm has a low hit rate in random workloads,resulting in performance degradation.Aiming at the problem of poor random read performance of traditional FTL algorithms in high-capacity SSDs,an FTL algorithm with high memory efficiency and flexible mapping is proposed,called Taichi-FTL.Taichi-FTL organizes consecutive logical pages into chunks,and maps them with chunk granularity.The mapping table is small and can be fully cached in DRAM to achieve fast address translation.At the same time,the logical pages randomly written in the chunk are managed with page granularity to ensure sufficient mapping flexibility.The two granular mappings can be switched freely according to the write size and there is no proportional limit,which completely eliminates the data migration overhead caused by merging.In addition,Taichi-FTL implements a high-capacity SSD mapping recovery optimization strategy to ensure rapid recovery of the mapping table.Taichi-FTL is implemented and evaluated on the FEMU SSD emulator.Experimental results show that compared with DFTL,Taichi-FTL can increase the average total throughput by 22.9%.While ensuring the write performance,the read request latency is reduced by 21.1% and the read bandwidth is increased by 18.9%.