Research On Performance Optimization Of Flash Translation Layer Based On 3D Flash Solid-state Drive

The performance of flash based Solid State Drive(SSD)is better than the traditional mechanical hard disk,and is gradually replacing the mechanical hard disk in reality.As flash technology continues to evolve,researchers have designed a new type of flash memory(3D flash)that is layered vertically on top of traditional planar(2D)flash memory.3D flash memory has higher storage density.In spite of this,the write performance of 3D flash memory is not stable.The main reason is that in the process of frequent write,due to the characteristics of non-overwrite write of flash memory,it will cause a large amount of write waste if the written data does not reach the size of a single flash page,especially for some load with small amount of write request data.In addition,garbage collection operations are frequently triggered,which affects the SSD life.To solve this problem,this paper proposes a Page Merge Flash Translation Layer(PMFTL)based on Page Merge.PMFTL proposes to add a part of 2D flash memory to 3D flash solid state disk,and make use of the unequal page sizes of two different flash media to process the write requests with different data volumes separately.The2 D flash memory in the data storage structure is used as the log area for receiving and transitioning write requests with small amount of data.3D flash memory as data area,as the final storage area of data.On the basis of the original global mapping directory table,a mapping table between log area and data area is established to build the mapping relationship between log area and data area.Different write back policies are used according to the data volume.When the storage capacity of the log area reaches the threshold for garbage collection,the data in the 2D flash valid data page is reorganized and then written back to the 3D flash data area.This strategy can effectively reduce write operations with small data volume in 3D flash memory,achieve performance optimization and prolong life.The experimental results show that the write performance is improved by 22.7% and 13.4% respectively compared with the pure page-level mapping and DFTL algorithms.Write magnification decreased by 30.2% and 21.1%.At the same time,this paper proposes a Garbage Collection strategy based on Double data-separate Garbage Collection(Ds GC).Ds GC separates the data twice,based on update frequency and update stability,with the ultimate goal of reducing the overhead of data migration in garbage collection.The first data separation occurs before data is written into flash memory.According to the number of data writes,data is divided into cold data and hot data and written into cold data and hot data respectively.The second data separation occurs in the process of data migration.By judging the stability of data update,data is divided into stable data and unstable data,which are stored in different physical blocks.In the process of garbage collection,the difference between the maximum erasable times between physical blocks,the proportion of invalid pages and the expiration time of invalid pages are taken into consideration.Simulation results show that compared with Greedy Algorithm(GR)and cost-benefit(CB)Algorithm,Ds GC Algorithm has a significantly lower number of block erasers and better wear balancing effect.