Research On Flash Translation Layer Of Solid State Drive

NAND flash based Solid State Drive(SSD) has been widely adopted in enterprise servers and high performance computing environments for its various advantages, such as low power consumption, high reliablity and high performance, etc. In order to make the interface of HDD compatible to SSD, solve the problems resulted from the internal characteristics of NAND Flash, such as erase-before-write and limited erase times, and improve the overall performance of SSD, Flash Translation Layer(FTL) is introduced between file system and NAND Flash memory to implement address mapping, garbage collection, and wear-leveling, etc. Address mapping focuses on translating logical address to physical address in flash, which can be classified into page-level, block-level and hybrid mapping depending on the granularity of address translation.Page-level mapping scheme is too expensive to utilize because of its large size of page mapping table. The mainstream DFTL is a very popular FTL scheme based on page-level mapping that selectively caches partial page-level address mappings in cache, which reduces the requirement of cache space. However, the heavy extra overheads introduced in DFTL affects the overall performance of SSD significantly.We propose an efficient cache management strategy of page-level address mappings, called ECAM that jointly exploits the temporal and spatial localities of workloads. ECAM records multiple address mappings in just one mapping entry with a SIZE attribute. In ECAM, the address mapping entries with large SIZEs have higher priority to stay in cache. ECAM manage the page-level address mappings in cache by exploiting the localities of workloads. Thus, ECAM can make use of the limited cache space efficiently, improve the cache hit ratio and reduce the extra overheads.We also propose a workload adaptive Flash Translation Layer algorithm with prediction based on ECAM, called WAPFTL, according to the internal characteristics of NAND flash that its performance is highly workload sensitive. WAPFTL can adjust its cache management strategy of address mappings dynamicly by predicting the I/O patterns of workloads. WAPFTL can solve cache space waste of address mapping information from random write requests, exploit the cache space more efficiently and get a further improvement of the overall performance of SSD.Our experiments show that both ECAM and WAPFTL can improve the cache hit ratio and reduce the extra overheads signifcantly, and improve the overall performance of SSD.