Optimal Design Of Solid State Disk Firmware Based On Hot Data Identification

With the advent of the information age,various information technologies have developed rapidly,leading to explosive growth in data.The storage and processing of these data make the memory face great challenges.Solid State Drive(SSD)takes full advantage of the high read and write performance of its flash storage media and has gradually replaced traditional mechanical hard drives.However,its inability to update in situ,limited lifespan,and asymmetry in read and write performance limit its full popularity.At present,SSD research has attracted great attention from domestic and foreign research scholars and has become a research hotspot in the storage field.This thesis conducts two studies based on the characteristics of cold and hot data and flash memory: 1)Based on the data access frequency and time locality,a hot data identification algorithm with high accuracy is proposed.2)Based on the hot data identification and the busyness of the channel queue,combined with the fast write depth characteristics of flash memory,a hybrid write strategy combining fast write and ordinary write is proposed to improve the performance of SSD.In the design of the hot data identification algorithm,combined with the characteristics of multiple real workloads,this thesis analyzes a variety of existing hot data identification algorithms.The existing algorithms do not comprehensively consider the issues of data access frequency and time locality.This thesis proposes a hot data identification algorithm based on Bloom filtering and temporal locality responding to the above problems,referred to as B2 L.The innovation is to first use the Bloom filter to initially identify the data.In this step,most of the cold data is removed,and only some data that are most likely to be hot are retained for subsequent identification.Then use the two-level LRU table to perform secondary identification on the obtained rough heat data to obtain the real heat data.B2 L not only avoids the problem of high false positives caused by pure Bloom filters due to hash conflicts but also effectively reduces the false positive rate of the secondary LRU identification algorithm and the false negative rate due to false positives,thereby greatly improving the hot data Identification accuracy.Experimental results show that compared with RSM,MIHF,MBF,DL-MBF,and 2LRU,B2 L has improved identification accuracy by 82.8%,63.9%,61.0%,47.6%,and 46.9%,respectively,without significant additional overhead.In the design of the hybrid write strategy,the traditional write strategy only determines the degree of channel blockage based on the length of the current flash memory channel request queue,thereby deciding whether to perform a fast write operation or a normal write operation.This strategy does not consider the hot and cold characteristics of the data,but in essence,it is also meaningful to use ordinary writing for cold data and fast writing for hot data for improving SSD performance.In response to the above problems,this thesis proposes a hybrid write strategy based on hot data identification and channel queue busyness,referred to as HBPHW.Its innovation lies in:on the one hand,it reflects the busyness of the channel according to the ratio of the current total number of sub-requests to be processed and the number of sub-requests that can be processed in parallel,and on the other hand,it reflects the heat of the data according to the hit distance of the request in the LRU table of B2 L.Then use the heat of data and channel busyness to make fast write decision modeling,and compare the result of the model with the fast write threshold to determine whether to perform fast write operations,to obtain high fast write benefits.In addition,HBPHW will adaptively adjust the fast write threshold according to the fast writing gains in a certain period recently,to maximize the fast writing gains.Experimental results show that compared with the case of no fast writing,HBPHW reduces the average response time by 54.27%,26.15%,40.68%,64.67%,61.94%,and 32.38%,respectively,without introducing excessive rewriting operations.