Research And Design Of Storage Management System For SSD

With the rapid development of the digital age,the total amount of global data transmission is increasing exponentially,and users have higher and higher requirements for access performance of storage devices.In this environment,flash-based Solid State Disk(SSD)are widely used due to their low latency and low power consumption,gradually replacing traditional hard disk driver.Although there have been some researches on SSD storage management systems,none of them can give full play to the potential of SSDs,and there are still a large number of technical problems to be solved urgently in SSD storage management systems.The storage management system consists of the upper I/O layer and the bottom device layer.In the SSD I/O layer,the traditional file system I/O stack is too thick.When applied to ultra-low-latency NVMe SSDs,the software overhead is far greater than the hardware overhead,which seriously affects the performance of the SSD.At the same time,the traditional file system does not consider the characteristics of flash memory,such as read-write asymmetry and write amplification,reduce the space utilization rate and endurance of SSD.In the device layer,the current mainstream Flash Translation Layer(FTL)algorithm usually loads part of the mapping records into DRAM when performing address mapping,while the complete address mapping table is persisted in logical address order in the flash memory,the hit rate of the Cache Mapping Table(CMT)is low,which affects the access performance of the SSD.In order to solve the above problems,this thesis designed a high-performance,high-reliability storage management system based on SSD.The main work and achievements of this thesis are as follows:1.Based on Intel’s SPDK technology,the N2 UFS file system based on NVMe SSD is implemented in the user space of the operating system.N2 UFS designs a more flexible space division scheme to improve the space utilization and random access performance of SSD.At the same time,the B-tree index technology is used to create a directory index tree in memory,which improves the metadata access performance of SSD.Aiming at the multi-queue feature of NVMe,an IO queue elastic scheduling algorithm is designed to improve the IOPS performance of SSDs.The experiments in this thesis show that N2 UFS improves throughput by 2.87% and random access performance by 32.06% compared with traditional file systems.2.A novel C-FTL algorithm based on thermal clustering idea is proposed,which improves the hit rate of CMT on the basis of DFTL.C-FTL centrally stores the recently eliminated mapping records from DRAM into the Hot Translation Page(HTP)of SSD,and generates its corresponding HTP index(HTPI)in DRAM for querying the HTP;at the same time,C-FTL also implements an auxiliary cache Sub-CMT in DRAM,which is used to optimize the organizational structure of HTP,reduce the false positive rate of HTPI,and further improve the efficiency of address translation.The experiments in this thesis show that compared with classic FTL algorithms such as DFTL,C-FTL has a 40.6%increase in the CMT hit rate,which improves the reliability of SSD access data;and CFTL does not increase the overhead of garbage collection,extending the endurance of SSD is shortened.