Research On The Firmware Optimization In Solid State Drive

With the rapid development of next-generation information technologies such as cloud computing,mobile Internet,and big data,efficient data storage and access has become a key issue to be urgently addressed in the storage field.After 2014,Flash-based solid-state drives(SSDs)began to explode and have a disruptive impact on the storage market.It has gradually become a research hotspot in the storage field.However,there are some limitations in SSDs,such as asymmetric read-write performance,erase before write,limited program/erase cycles and not be directly applied to traditional file systems.In addition,the latest Flash manufacturing technology(such as three-dimensional stacked TLC flash memory)makes the storage density of SSD further improve and the unit capacity cost continue to decline,but it also posed a higher challenge to the performance and life of SSD.In the design of the flash translation layer(FTL),the existing DFTL and its improved algorithms fail to solve the following three problems: 1)They cannot make full use of the read and write characteristics of traces to optimize FTL;2)Their write-back mechanism for dirty mapping entries is insufficient,resulting in frequent translation page updating;3)The hot and cold data is not distinguished,resulting in low garbage collection efficiency.Aiming at the above problems,this paper proposes an adaptive read-write partitioning flash translation layer algorithm(ARWFTL).The contributions of ARWFTL are as follows: 1)A read-write separated mapping cache table(namely,R-CMT and W-CMT)is proposed,and the size of the two tables can be dynamically adjusted according to their unit benefit,which can maximize the mapping cache space utilization.2)A novel Clean entry First and Clustering based Least Recently Used(CFC-LRU)strategy is proposed to reduce the number of the translation page update significantly.3)A hot data window is set at the head of W-CMT and a hot-cold aware data write strategy is proposed to solve the hot-cold data entanglement problem in the underlying flash data blocks,which can reduce the number of valid pages migration in garbage collection.The experimental results show that compared with DFTL,CPFTL,HCFTL,and IRRFTL,ARWFTL reduces the number of translation page writeback by 92.8%,80.2%,73.9% and 56.3%,the number of valid page migrations by 47.7%,18.9%,13.8%,and 2.8%,the number of block erases by 31.6%,13.4%,9.1%,and 5.7%,and improves response time by 23.0%,15.9%,6.2% and 6.0%,respectively.By investigating the characteristics of flash memory,this paper finds that there is a trade-off between data write speed and data retention time in flash memory.Specifically,the write speed of flash memory can be increased by shortening the data retention time to improve the performance of SSD.Based on this purpose,this paper proposes a Hot Data Aware Fast Write Management strategy(HDAFWM).Combining the characteristics of short retention time of hot write data,HDAFWM adopts a fast write strategy for hot data and normal write strategy for cold data,so as to further improve the performance of FTL.The contributions of HDAFWM are as follows: 1)Divide the flash block into hot and cold blocks to store hot data using fast write strategy and cold data using normal write strategy;2)Use the first-in-first-out policy to manage the write queue to ensure that the valid data in the hot write block will not be invalidated due to the expiration of the retention time;3)By perceiving the benefit of fast write and the cost of rewrite,the amount of hot data identified is adjusted to achieve the tradeoff between the minimum cost of rewrite and the maximum benefit of fast write.The experimental results show that compared with ARWFTL,the average response time of ARWFTL based on HDAFWM is reduced by about 22% without significantly increasing the number of block erase,which verifies the effectiveness of the proposed HDAFWM strategy.