Research On The Hybrid Address Mapping Algorithm Of Flash Memory

The non-volatile, low power and high stability and so on characteristics of Flashmemory make it become one of the best storage media. However, Flash memory hassome inherent limitations such as limited-erase-cycles, erase-before-write, andasymmetric access latency between read and write operations. So the storage systemequips a software layer, called Flash Translation Layer (FTL), which include addressmapping, garbage collection and wear leveling. We focus on the address mappingstrategy which is the most basic and critical in this work. And we propose a novelhybrid NAND Flash address mapping management strategy on the basis of researchingthe classical hybrid address mapping algorithms.The classical hybrid address mapping mechanism encounter poor random writeperformance generally. Considering this issue we improve the mapping table structureof hybrid address mapping mechanism. First, we add remaining page number and thelog block list information to the conventional block-level address mapping table. So itcan call the log block adaptively according to the characteristics of the workloads.Second, all of the page-level address mapping entries are stored in the NAND Flash.Dynamically load and unload page-level mapping information to the SRAM by makingfull use of temporal locality and spatial locality of the workloads. Such hybrid addressmapping strategy can significantly reduce the full merge operations of the garbagecollection, and effectively delay the garbage collection operation. Thereby we canimprove the random write performance of Flash memory, so we call our algorithm asFRW (Fast Read and Write) for the meaning of quick access.We compared our optimization algorithm FRW with the classical hybrid FASTalgorithm on FlashSim simulator with a series of trace files being workloads, FRWoptimization algorithm has a better performance in write access speed, garbagecollection overhead, and the overall performance in different degrees. Among them,when the OLTP trace which has the characteristics of random access is the workloads,the improvement ratio of the average response time of the system can reach above30%.In order to test our FRW more comprehensively, we further test FRW optimizationalgorithm on the NAND Flash hardware platform of HITATCI compared with FAST insequential and random access performance. Since the sequential access performance ofFAST algorithm has been relatively satisfactory, the sequential access performanceof FRW algorithm is almost the same with FAST. However, the improvement of FRWalgorithm in random access performance is ideal. The improvement ratio is up toapproximately34.6%when the size of the transfer files is different.