Research On A Kind Of Flash Memory-oriented Temperature Data Perception Algorithm And Application

With the development of big data,high performance computing is becoming more and more important in many fields.HDD are increasingly unable to meet the actual demand for high speed I/O.SSD based NAND Flash have been widely used in many applications,but Flash has many inherent characteristics,for example:I/O asymmetry,pre-write erase,out-of-place updates,limited lifetime and so on.These feature will generate many invalid and dead pages,which can degrade efficiency and performance.Now,in order to improve the efficiency of garbage collection and the degree of wear leveling,an effective method that can be used is hot data identification.In this thesis,in order to improve the current disadvantages of hot and cold identification algorithm,an effective algorithm that can be proposed is data temperature perception algorithm.It divides the original data access cycle into multiple sub-cycles,and through CBF to record the number of times of statistical page requests in each sub-cycle,and use BF to record the data page of reaching the threshold in each sub-cycle,while the temperature concept can be quoted,it is use warm,hot and cold state to represent data access.By dong so,it can identify the occasional heating of occasional cold data,Therefore,it can further improve the accuracy of the data page recognition,through theoretical and practical analysis,its have a good recognition accuracy,lower time and space complexity.In this thesis,the data temperature perception algorithm is applied to the cache management of SSD.In view of the characteristics of flash memory,it is proposed to write the data pages that are replaced in the buffer area into the same flash block according to the same access rules,therefore,it can reduce the write magnification and improving the efficiency of garbage collection.The results show that the algorithm is improve 32%of the recognition accuracy,save nearly 23%of the memory consumption and increased by nearly 13%of the write efficiency by comparing with MBF algorithm in terms of recognition accuracy,memory consumption and number of block erasure,and the number of effective pages in block erase in the SSDSim simulator.Due to the inherent characteristics and price factors of the SSD,it is not possible to completely replace the HDD in the future.Therefore,combining the advantages of SSD and HDD is the future research direction.Based on the improvement of the time-sensitive hybrid storage model,Data temperature perception hybrid storage model,that is,the use of data temperature perception algorithm to carry out a high degree of accuracy of the page to identify the data to reduce the mobility of the overall I/O efficiency.By compared with the time-sensitive model in terms of the run time and data migration cost,the experimental results show that the proposed model has good performance,which can improve the overall I/O performance and reduce the unnecessary data migration.