Research On Key Technologies Of Parallel Design Of Solid State Drives

Efficient data storage and access have become the key problem to be solved in the storage field with the rapid development of 5G wireless communication,cloud computing,big data,and other emerging information technologies.Solid state drive(SSD)with high-speed read-write performance has become one of the mainstream storage devices.However,there are some limitations in SSD flash media,such as asymmetric read-write performance,out-of-place update,and limited program/erase recycles.How to design the firmware algorithm according to the characteristics of flash memory has become a research hotspot in the SSD field.This thesis solves two problems based on the parallel structure of the underlying flash memory: 1)improving the read-write parallelism of flash memory through IO scheduling and achieving wear-leveling at the scheduling level;2)reducing the impact of garbage collection on host IO,improve wear-leveling,and maximize garbage collection efficiency through the combination of interruptible garbage collection and non-interruptible garbage collection.The existing IO schedulers fail to solve the following two problems: 1)the mapping information of FTL is not fully utilized;2)the wear degree of underlying flash memory is not fully considered when allocating write requests.To solve these problems,the wear-aware out-of-order dynamic scheduling algorithm(WODSA)is proposed in this thesis.First,according to the mapping information of the flash translation layer,the max-parallel-based scheduling strategy is proposed to schedule the read requests to maximize read parallelism and reduce the waiting time.Second,for write requests,the wear-aware dynamic write allocation strategy is proposed based on the idle/busy state and wear degree of channels and chips.WODSA allocates write requests to channels and chips with less wear preferentially in a maximized parallel manner to achieve write parallelism and active dynamic wear-leveling.Compared with PAQ and PIQ,the average response time of WODSA is reduced by 17.2% and 22.7%,and the max waiting time is reduced by 60.1% and 53.5%,respectively.Besides,WODSA has the best wear-leveling performance.The existing garbage collection algorithms fail to solve the following problems: 1)the garbage collection IO and host IO are not uniformly scheduled which will cause performance fluctuation of SSD;2)how to choose the time to schedule garbage collection IO without blocking subsequent host IO;3)how to make a compromise between the garbage collection efficiency and wear-leveling when selecting the victim block.To solve these problems,this thesis proposes the uniform scheduling of interruptible garbage collection(USIGC).First,USIGC sets up an interruptible garbage collection sub-request queue and then schedules the sub-request queue with the host IO queue.In this way,the idle time of each channel is fully utilized to complete the valid page migration and erase operation of interruptible garbage collection.Second,USIGC predicts the probability that the erase operation can be completed in the current idle time based on the historical idle interval and then makes the erase operation decision based on this probability to achieve not blocking the host IO.Third,by converting the amount of data that can be written in the future to the present and taking the number of invalid pages of the current block as the selection basis of the victim block,the garbage collection and wear-leveling can be unified.Compared with DTGC and FAGC+,the average response time of USIGC is reduced by 8.4% and 13.2%,the max waiting time is reduced by 9.2% and 16.9%,the block erase standard deviation is reduced by 12.5% and 12.2%,and the channel erase standard deviation is reduced by 62.3% and 32.1%,respectively.