| In the information age,the amount of data generated by the world each year exceeds 33 ZB.As a resident of data,the storage device’s reliability and security are directly related to national information security and economic operation security.Therefore,it has become an urgent need to study high-speed and large capacity reliable storage technology and develop high-reliability large capacity storage systems,which has important strategic significance.Unlike Hard-Disk Drive(HDD),the Solid-State Drive(SSD)is a completely new storage method based on semiconductors.It stores data by changing the state of charge,phase change,or polarity of the storage medium,so it has high reliability,high performance,and low power consumption.And significant advantages such as non-volatile,is currently gradually replacing HDD as the main data storage device.However,the storage principle and manufacturing process of semiconductor storage media determine the inherent problems such as limited SSD life and reduced reliability due to limited erasure times.The purpose of this thesis is to improve the reliability of data storage.Combined with the characteristics of NAND Flash,this thesis studies several key technologies for optimizing the storage of SSD data,such as the improved algorithm of wear leveling,off-line data deduplication method,small-scale parallel compression algorithm and multi-channel data redundancy check method based on hot and cold strips,which not only improves the reliability of SSD,but also ensures that the read-write performance of SSD does not decrease basically.The main innovative research work of this thesis is as follows:1.In this study,a wear leveling method based on page-level fine division was designed.The data stored in an SSD was divided into cold and warm states according to the data update times.By comparing the update times of each data page in the warm block,the data with the most frequent update times was identified as hot data.Combine these hot data into a block to make the block a real hot block,and perform the data in the real hot block.The data in the real hot block and the cold block were exchanged in storage locations.Simulation results show that this method could identify the thermal data in SSD more accurately,and improved the reliability of SSD without affecting the performance of SSD.At present,this method has been applied to the task data loading and unloading system of a new transport aircraft.2.The SSD data deduplication method can effectively reduce the amount of data written into NAND Flash and fundamentally reduce the wear of NAND Flash.Typical inline SSD data deduplication methods mostly use FPGA to accelerate in order to ensure performance,resulting in complex system and increased cost.Based on the existing embedded memory controller,this study designed an off-line SSD data deduplication method without FPGA acceleration.In this method,the corresponding fingerprints were generated for the cold data identified by the improved wear leveling algorithm proposed in this thesis,all fingerprints were compared and the cold data corresponding to the same fingerprint was deleted,and then the cold data and hot data after deduplication were exchanged.The simulation results show that the average repeated data recognition rate of this method is 22%,which is close to the recognition rate of the complex and expensive online data de-duplication method.Compared with the SSD without data deduplication function,the performance of the SSD using this method was improved by 15%,and the performance was similar to that of the SSD using inline data deduplication.3.In order to further to reduce the amount of data written to SSD,this thesis designed a smallscale parallel compression method suitable for SSD based on LZW data compression calculation method.Firstly,the storage space of SSD was divided into a hot area and a plurality of cold areas according to the erasing frequency,and then the data blocks in each area were divided into hot data blocks and cold data blocks according to the erasing times,and the address mapping table of the hot area was added to the cache to improve the reading and writing performance of data in the hot area.When data was written,if the data belongs to cold data,compress the data immediately.If the data belongs to cold data,the data would be stored in cache,and then compressed when the hot area in cache needs to be written back to NAND Flash.By dividing the storage space of SSD into cold and hot areas,unnecessary data compression work was reduced and data compression efficiency was improved.The test results shown that the data compression method designed in this study could effectively reduce the write amplification factor of SSD,and could effectively improve the reliability and performance of SSD.4.In this study a method for redundant stripe data redundancy check in SSD was proposed.In this method,the NAND Flash memory channel arrays of SSD were divided into two independent areas,one was the data storage area of user data and the other was the parity storage area of parity.Each storage area was divided into several stripes according to the parallel relationship of storage channels,and two independent flash translation layers were designed,which were used for the storage management of the two storage areas respectively.Because the data in the parity area may need to be updated every time it is accessed,it belongs to the weakest area of SSD life.According to the update times of each stripe in the storage area,this study divided it into cold stripes and hot stripes.The parities corresponding to cold stripes could be directly written into NAND Flash due to relatively few updates,while the parities corresponding to hot stripes were frequently updated,which would seriously affect the life of SSD.Therefore,this study created a data buffer in memory to store the parities corresponding to the hot stripes,and generated and write the parities in a time-sharing manner.When the hot stripes become the cold stripes,the parities in the buffer were written into NAND Flash,which avoided the excessive wear of NAND Flash caused by frequent writing of parities.Simulation results shown that compared with RAID-5,which was the most commonly used array verification method,this method could not only reduce the write amplification factor of SSD,but also improved the performance of SSD.5.At present,the research on key technologies of SSD storage is mainly verified and tested by SSD simulator.Although the SSD simulator has been able to simulate and test various key technologies,these technologies are to be used in actual storage products,they must be further verified and tested by the SSD development platform.The existing SSD development platform has single interface function and limited hardware resources.So it is impossible to comprehensively verify all the methods designed in the thesis on one platform.Aiming at the deficiency of the existing SSD development platform,this thesis designed a new generation SSD development platform with dynamic reconfigurable function.The platform is composed of several FPGAs,and a variety of mainstream SSD storage interfaces were designed,which can verify the current cutting-edge data of storage technology and provide hardware support for further research of SSD storage technology. |
PDF Full Text Request