Solid state disk cold and hot data intelligent layered storage method and system

Abstract

The application relates to the technical field of data storage, and discloses a solid state disk cold and hot data intelligent layered storage method and system.The method comprises the following steps: collecting spatial position change data and vibration state data, and obtaining a multidimensional feature data set after preprocessing; quantitatively processing the influence degree of the spatial position change data and the vibration state data to obtain an interference weight value; constructing a joint feature matrix based on the interference weight value and historical access records to predict hot spot data distribution; if the matching degree of the hot spot data distribution and a current storage layout is lower than a preset matching degree threshold, migrating high-frequency access data to a low-latency area to obtain an optimized storage layout; when an access response time is greater than a preset time threshold, generating log data and extracting delay load features; and secondarily adjusting the optimized storage layout to obtain a final dynamic storage scheme.The method can realize dynamic optimization of the storage performance of a solid state disk and reduce delay loss caused by cold and hot data mismatch.

Description

Technical Field

[0001] This invention relates to the field of data storage technology, and in particular to a method and system for intelligent hierarchical storage of cold and hot data on a solid-state drive. Background Technology

[0002] In modern information technology fields such as cloud computing, big data processing, and artificial intelligence, solid-state drives (SSDs) have become the core carrier of data storage due to their advantages such as high-speed read / write and low power consumption. Their storage performance directly determines the operating efficiency of the entire system. With the exponential growth of data volume across various fields and the diversification and randomization of data access patterns, how to achieve efficient hierarchical data management to match different access needs has become a core issue in improving the utilization efficiency of SSDs.

[0003] The mainstream approach to tiered data management for solid-state drives (SSDs) involves classifying and storing data using pre-defined fixed rules or simple caching mechanisms. Data is categorized into hot and cold categories based on a unified set of criteria and allocated to corresponding storage areas. Recently accessed data is then temporarily cached to reduce disk reads of duplicate data. However, as SSD applications expand, the dynamic nature of their operating environment becomes increasingly apparent. Existing methods fail to incorporate environmental factors into their data management decision-making logic, causing the tiered rules, based on a static environment, to become disconnected from actual data access needs. This makes it impossible to predict changes in data access patterns based on environmental changes, resulting in system response speeds that cannot meet low-latency requirements.

[0004] In summary, existing solid-state drive (SSD) data tiering management methods struggle to incorporate adaptive decision-making logic based on dynamic environmental factors, resulting in insufficient response speed in highly dynamic scenarios. Summary of the Invention

[0005] This invention provides a method and system for intelligent tiered storage of cold and hot data in solid-state drives (SSDs), which enables dynamic optimization of SSD storage performance under complex operating environments and improves the overall system response speed.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for intelligent tiered storage of hot and cold data in a solid-state drive, comprising:

[0007] Collect spatial position change data and vibration state data of solid-state drives during operation, perform preprocessing, and obtain a multidimensional feature dataset;

[0008] Based on the multidimensional feature dataset, the efficiency impact of the spatial location change data and the vibration state data is quantified to obtain the corresponding interference weight value.

[0009] Based on the interference weight values ​​and the pre-acquired historical access records, a joint feature matrix is ​​constructed, and based on the joint feature matrix, a long short-term memory network model is used to predict the distribution of hot data within a future preset time period.

[0010] Obtain the current storage layout, calculate the matching degree between the hot data distribution and the current storage layout, and if the matching degree is lower than a preset matching degree threshold, migrate the high-frequency access data to a low-latency area to obtain an optimized storage layout.

[0011] Based on the optimized storage layout, the response time of data access requests is monitored in real time. When the response time is greater than a preset time threshold, log data containing latency feedback is generated, and latency load characteristics are extracted.

[0012] In response to the aforementioned latency load characteristics, the optimized storage layout is further adjusted by adjusting the data migration frequency and priority to obtain the final dynamic storage solution.

[0013] Preferably, the step of quantifying the efficiency impact of the spatial location change data and the vibration state data based on the multidimensional feature dataset to obtain corresponding interference weight values ​​includes:

[0014] Based on the multidimensional feature dataset, the spatial coordinate offset corresponding to the spatial position change data is extracted, and time series analysis is performed on the vibration state data to obtain the corresponding vibration frequency data.

[0015] The spatial coordinate offset is input into a pre-trained offset-interference regression model to obtain the position interference level, and the position interference factor is determined based on the position interference level.

[0016] The vibration interference factor is obtained by matching the preset vibration interference mapping relationship with the vibration frequency data;

[0017] The position interference factor and the vibration interference factor are nonlinearly combined to obtain a comprehensive interference weight value.

[0018] Preferably, the step of constructing a joint feature matrix based on the interference weight values ​​and pre-acquired historical access records includes:

[0019] Extract data access frequency from the historical access records. If the data access frequency exceeds a preset frequency threshold range, it is marked as abnormal behavior, and behavioral feature data is obtained.

[0020] Obtain the current environmental fluctuation parameters, and combine them with the behavioral feature data to identify the interfering factors that cause a decrease in access efficiency, and determine them as environmental interference features;

[0021] Based on the environmental interference characteristics, when the interference weight value is greater than the preset weight threshold, the environmental fluctuation parameter is adaptively adjusted to obtain the parameter adjustment result;

[0022] The parameter adjustment results, environmental interference features, and behavioral feature data are fused in multiple dimensions to obtain the joint feature matrix.

[0023] Preferably, the step of predicting the distribution of hotspot data within a preset time period using a long short-term memory network model based on the joint feature matrix includes:

[0024] Historical access frequency sequences, environmental parameter sequences, and delay data sequences of different types of data are extracted from the joint feature matrix and combined to construct a time series feature set.

[0025] The time series feature set is input into the pre-trained long short-term memory network model to obtain the temporal correlation features between access patterns and environmental factors;

[0026] Based on the aforementioned temporal correlation features, predict the trend of changes in the access frequency of different types of data within a future preset time period;

[0027] Based on the trend of access frequency changes, data blocks with access frequencies higher than a preset hotspot threshold are identified as the hotspot data distribution.

[0028] Preferably, the step of calculating the matching degree between the hot data distribution and the current storage layout, and if the matching degree is lower than a preset matching degree threshold, then migrating high-frequency access data to a low-latency area to obtain an optimized storage layout, includes:

[0029] Scan the storage metadata in the current storage layout, compare the storage metadata with the hot data distribution, calculate the proportion of hot data located in the high-latency area, and obtain the matching degree;

[0030] When the matching degree is lower than the preset matching degree threshold, high-frequency access data located in the high-latency region is selected from the hot data distribution and determined as the data set to be adjusted.

[0031] The data set to be adjusted is prioritized according to access frequency from high to low, and a migration target list is generated by combining the real-time storage capacity information of the low-latency area.

[0032] According to the migration target list, the high-frequency access data is migrated to the low-latency area in batches to obtain the storage layout after migration;

[0033] Calculate the matching degree between the migrated storage layout and the hot data distribution. If it is still lower than the preset matching degree threshold, adjust the data layering rules to obtain an optimized storage layout.

[0034] Preferably, the step of monitoring the response time of data access requests in real time according to the optimized storage layout, and generating log data containing latency feedback when the response time exceeds a preset time threshold, and extracting latency load characteristics, includes:

[0035] The response time of data access requests is continuously collected from the optimized storage layout;

[0036] When the response time exceeds a preset time threshold, log data containing latency feedback is generated, and access load fluctuation information is extracted from the log data.

[0037] When the fluctuation information exceeds a preset fluctuation threshold, the load change pattern in the fluctuation information is analyzed, and the delay load characteristics are obtained by combining the storage area information where the delay occurs.

[0038] Preferably, the step of adjusting the optimized storage layout by adjusting the data migration frequency and priority based on the latency load characteristics to obtain the final dynamic storage solution includes:

[0039] Based on the latency load characteristics, high-frequency latency data and their corresponding latency anomaly regions are identified, and a list of data to be migrated is generated.

[0040] Obtain the remaining capacity of the current low-latency region, and combine it with the list of data to be migrated and the remaining capacity to determine the migration frequency table of the high-frequency latency data, thus obtaining the migration frequency table;

[0041] Based on the migration frequency table, the high-frequency latency data is prioritized and sorted from high to low access frequency to obtain a secondary migration target list.

[0042] Based on the secondary migration target list, the high-frequency latency data is migrated to the low-latency area in batches according to priority, resulting in the final dynamic storage scheme.

[0043] Secondly, this invention provides a smart tiered storage system for cold and hot data in a solid-state drive, comprising:

[0044] The data acquisition and preprocessing module is used to collect spatial position change data and vibration state data of the solid-state drive during operation, perform preprocessing, and obtain a multi-dimensional feature dataset.

[0045] The interference weight quantization module is used to quantify the degree of efficiency impact of the spatial position change data and the vibration state data based on the multidimensional feature dataset, and obtain the corresponding interference weight value.

[0046] The feature matrix construction module is used to construct a joint feature matrix based on the interference weight values ​​and pre-acquired historical access records;

[0047] The hotspot data prediction module is used to predict the distribution of hotspot data within a preset time period based on the joint feature matrix and a long short-term memory network model.

[0048] The storage layout optimization module is used to obtain the current storage layout, calculate the matching degree between the hot data distribution and the current storage layout, and if the matching degree is lower than a preset matching degree threshold, then the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout.

[0049] The latency load extraction module is used to monitor the response time of data access requests in real time according to the optimized storage layout. When the response time is greater than a preset time threshold, log data containing latency feedback is generated and latency load characteristics are extracted.

[0050] The dynamic scheme adjustment module is used to make secondary adjustments to the optimized storage layout based on the latency load characteristics by adjusting the data migration frequency and priority, so as to obtain the final dynamic storage scheme.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) This invention obtains a multidimensional feature dataset by collecting spatial position change data and vibration state data of a solid-state drive during operation and preprocessing them. Then, it quantifies the degree of influence of this multidimensional feature dataset on the efficiency of the two types of data to obtain interference weight values. This process can transform the dynamic environmental factors in the operation of the solid-state drive into quantifiable analytical indicators, enabling the system to accurately grasp the specific degree of influence of spatial position change and vibration state on data access efficiency, thereby improving the adaptability of data hierarchical management to the dynamic environment and ensuring that the data management strategy always remains coordinated with the current environmental state.

[0053] (2) This invention constructs a joint feature matrix based on interference weights and pre-acquired historical access records, and then uses a long short-term memory network model to predict the distribution of hot data within a preset time period based on the joint feature matrix. The joint feature matrix integrates dynamic environmental interference information and historical data access patterns, while the long short-term memory network model has the advantage of capturing long-term dependencies in time series data. The combination of the two can accurately uncover the potential patterns of data access patterns changing with the environment, and reduce efficiency losses caused by the mismatch between data storage location and access needs.

[0054] (3) This invention obtains the current storage layout and calculates the matching degree between the distribution of hot data and the layout. If the matching degree is lower than a preset matching degree threshold, the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout. By judging the matching degree, the mismatch between the storage layout and the demand for hot data can be accurately located. Migrating the high-frequency access data to a low-latency area can directly shorten the access path of high-frequency data, enabling the system to respond quickly when processing high-frequency access requests, significantly improving the access efficiency of high-frequency data, and reducing the access latency of core data.

[0055] (4) This invention monitors the response time of data access requests in real time based on the optimized storage layout. When the response time exceeds a preset time threshold, log data containing latency feedback is generated and latency load characteristics are extracted. Then, by adjusting the data migration frequency and priority, the optimized storage layout is adjusted a second time to obtain the final dynamic storage solution. Real-time monitoring of response time can promptly capture efficiency deviations in the actual operation of the storage layout. Extracting latency load characteristics can accurately locate the load-level causes of latency. The secondary adjustment based on these characteristics enables the storage layout to dynamically adapt to changes in the current access load and environmental conditions, ensuring that the solid-state drive maintains stable and efficient storage performance over a long period of time. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating an embodiment of the intelligent hierarchical storage method for cold and hot data in solid-state drives provided by the present invention.

[0057] Figure 2 This is a schematic diagram of an embodiment of the intelligent tiered storage system for cold and hot data in solid-state drives provided by the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Reference Figure 1 The first embodiment of the present invention provides a method for intelligent tiered storage of hot and cold data in a solid-state drive, including steps S11 to S17, as follows:

[0060] S11: Collect spatial position change data and vibration status data of the solid-state drive during operation, perform preprocessing, and obtain a multi-dimensional feature dataset;

[0061] S12, based on the multidimensional feature dataset, quantify the degree of efficiency impact of the spatial position change data and the vibration state data to obtain the corresponding interference weight value;

[0062] S13, construct a joint feature matrix based on the interference weight value and the pre-acquired historical access records;

[0063] S14. Based on the joint feature matrix, a long short-term memory network model is used to predict the distribution of hot data within a preset time period in the future.

[0064] S15, obtain the current storage layout, calculate the matching degree between the hot data distribution and the current storage layout, if the matching degree is lower than the preset matching degree threshold, then migrate the high-frequency access data to the low-latency area to obtain an optimized storage layout.

[0065] S16. Based on the optimized storage layout, monitor the response time of data access requests in real time. When the response time is greater than a preset time threshold, generate log data containing latency feedback and extract latency load characteristics.

[0066] S17. Based on the latency load characteristics, the optimized storage layout is adjusted a second time by adjusting the data migration frequency and priority to obtain the final dynamic storage solution.

[0067] In step S11, spatial position change data and vibration state data of the solid-state drive during operation are collected, preprocessed, and a multidimensional feature dataset is obtained.

[0068] It's important to note that spatial position change data refers to the real-time coordinate information of the solid-state drive (SSD) in three-dimensional physical space. This data records the SSD's position coordinates in the x, y, and z dimensions. High-precision position sensors track position movement to determine the impact of changes in the SSD's physical orientation on the relative position of the read / write head and storage units. For example, excessive positional offset can alter the read / write path length, thereby increasing data access latency. Vibration status data refers to the number of vibrations the SSD experiences per unit time. This data quantifies the interference of environmental vibrations on the SSD's hardware stability and can be obtained in real-time using an accelerometer. Excessively high vibration frequencies can lead to unstable contact between the read / write head and storage units, causing data read errors or latency fluctuations.

[0069] In this embodiment, the preprocessing step aims to eliminate interference information in the original data, fill data gaps, and uncover patterns of environmental changes. Since the original data is susceptible to interference from environmental factors such as fan rotation and equipment heat dissipation in the computer room, noise reduction is first required. This is achieved by performing an equal-weighted average calculation on the original data from five consecutive time points, reducing the impact of random noise in a single data point and preserving the true trend of the data's changes. Next, for missing data values, linear interpolation is used to estimate the missing values ​​based on valid data from adjacent time points before and after the missing data, thus obtaining a complete dataset. For example, if the z-axis coordinate is 5.1 cm at time t1 and 5.3 cm at time t3, then the estimated missing z-axis coordinate at time t2 (the midpoint between t1 and t3) is 5.2 cm.

[0070] Furthermore, based on the denoised complete dataset, the fluctuation ranges of both location coordinates and vibration frequency are statistically analyzed to obtain preliminary descriptive information about the dynamic changes in the environment. For example, the daily average fluctuation range of vibration frequency is 20 times / s, and the daily average change range of location coordinates is 0.5cm. Simultaneously, the time points of abnormal fluctuations and their corresponding hard drive spatial locations are located from the preliminary descriptive information. For instance, if at 10:03 AM on a certain day, the hard drive vibration frequency suddenly rises to 80 times / s, and if the preset abnormal vibration frequency threshold is 70 times / s, an alarm mechanism is triggered to record the specific time of the abnormal event as 10:03 AM, and the hard drive location coordinates as x=10.5cm, y=20.3cm, z=5.2cm. This abnormal data is then archived in the abnormal database. Finally, by integrating the denoised and completed location coordinate data, vibration frequency data, and fluctuation ranges obtained from time series analysis, combined with the archived abnormal fluctuation information, a multi-dimensional feature dataset containing multi-dimensional information is constructed. This dataset comprehensively reflects the dynamic characteristics of the solid-state drive's operating environment, improving the adaptability and effectiveness of the data hierarchical management strategy.

[0071] In step S12, based on the multidimensional feature dataset, the efficiency impact of the spatial location change data and the vibration state data is quantified to obtain the corresponding interference weight values, including:

[0072] Based on the multidimensional feature dataset, the spatial coordinate offset corresponding to the spatial position change data is extracted, and time series analysis is performed on the vibration state data to obtain the corresponding vibration frequency data.

[0073] The spatial coordinate offset is input into a pre-trained offset-interference regression model to obtain the position interference level, and the position interference factor is determined based on the position interference level.

[0074] The vibration interference factor is obtained by matching the preset vibration interference mapping relationship with the vibration frequency data;

[0075] The position interference factor and the vibration interference factor are nonlinearly combined to obtain a comprehensive interference weight value.

[0076] It's important to note that spatial coordinate offset refers to the change in the three-dimensional coordinates of the solid-state drive (SSD) within a preset time interval. It's calculated by taking the Euclidean distance between the coordinates of two consecutive points in time, reflecting the magnitude of the physical position shift of the drive. The preset time interval is determined based on the temporal characteristics of SSD data access and the efficiency optimization requirements of real-world application scenarios. It ensures that the spatial coordinate offset accurately reflects the impact of environmental changes on data access efficiency. A larger offset indicates a more significant deviation in the relative position between the read / write head and the target storage unit, leading to a longer read / write path and increased data access latency. Vibration frequency data refers to the number of vibrations the SSD experiences per unit time, quantifying the intensity of vibration interference with hardware stability. A higher vibration frequency results in poorer contact stability between the read / write head and the storage unit, easily causing signal transmission fluctuations, leading to increased access latency or read errors.

[0077] In this embodiment, the offset-interference regression model is a linear regression model, taking the spatial coordinate offset x as input and the corresponding delay increase ratio R as output. Mapping is achieved through the linear function R = k × x + b, where k is the proportionality coefficient (unit: % / cm) and b is the intercept (unit: %). Based on historical measured data from the past 30 days, a total of 1000 samples were collected. Using the least squares method, k and b were calculated by minimizing the mean square error between the predicted and actual delay increase ratios to ensure model fitting accuracy. Then, the location interference levels are classified according to the delay increase ratio output by the model. According to the preset level-factor mapping rule, the levels are converted into location interference factors between 0 and 1. The factor values ​​satisfy the monotonicity of "the higher the level, the larger the factor," and the factor difference between adjacent levels matches the increase in the delay increase ratio. For example, when the offset is 0-0.2cm, the delay increases by ≤5%, the interference level is 1, and the corresponding factor is 0.1; when the offset is 0.2-0.4cm, the delay increases by 5%-10%, the interference level is 2, and the corresponding factor is 0.3; when the offset is 0.4-0.6cm, the delay increases by 10%-20%, the interference level is 3, and the corresponding factor is 0.5.

[0078] Furthermore, the preset vibration interference mapping relationship is constructed based on a large amount of experimental data. By testing the access latency changes of solid-state drives under different vibration frequencies, the correlation between the latency increase ratio and vibration frequency is statistically analyzed, and frequency intervals are divided and corresponding interference factors are assigned. For example, when the vibration frequency is 0-30 times / s, the latency increase is ≤5%, corresponding to a vibration interference factor of 0.1. Both interference factors are used to quantify the degree of environmental interference, and both range from 0 to 1. The larger the factor value, the stronger the interference of the corresponding environmental change on access efficiency.

[0079] Furthermore, since spatial location changes and vibrations do not independently affect access efficiency, when both are at high levels of disturbance, the combined increase in delay caused by both is greater than the sum of the effects of a single factor. Therefore, this synergistic relationship needs to be reflected through a nonlinear formula. In this method, a comprehensive disturbance weight is used. The calculation formula is: ,in and These represent the position disturbance factor and the vibration disturbance factor, respectively. and These represent the independent interference weight percentages for position and vibration, respectively, while the other two represent the combined interference weight percentages for both. and This item is used to enhance the influence of high interference factors. The item is used to reflect the synergistic effect. The interference weight ranges from 0 to 1, with a larger value indicating a stronger overall interference from environmental factors on data access efficiency.

[0080] For example, the spatial coordinate offset of 0.5cm and vibration frequency data of 60 times / s within 10 minutes were extracted from a solid-state drive in a server room. First, the 0.5cm spatial coordinate offset was input into the offset-interference regression model. According to the level standard of the model, the output position interference level was 3. Then, the position interference factor P=0.5 was obtained through the level-factor mapping rule. Next, the vibration frequency data of 60 times / s was matched with the preset vibration interference mapping relationship, and the corresponding vibration interference factor V=0.6 was obtained. Finally, the interference weight ratios α=0.4, β=0.4, and γ=0.2 were set and substituted into the nonlinear combination formula to calculate the comprehensive interference weight value W=0.4×(0.5)²+0.4×(0.6)²+0.2×0.5×0.6=0.1+0.144+0.06=0.304, indicating that the comprehensive interference of the current environment on the access efficiency of the solid-state drive is at a low to medium level.

[0081] In step S13, a joint feature matrix is ​​constructed based on the interference weight value and the pre-acquired historical access records, including:

[0082] Extract data access frequency from the historical access records. If the data access frequency exceeds a preset frequency threshold range, it is marked as abnormal behavior, and behavioral feature data is obtained.

[0083] Obtain the current environmental fluctuation parameters, and combine them with the behavioral feature data to identify the interfering factors that cause a decrease in access efficiency, and determine them as environmental interference features;

[0084] Based on the environmental interference characteristics, when the interference weight value is greater than the preset weight threshold, the environmental fluctuation parameter is adaptively adjusted to obtain the parameter adjustment result;

[0085] The parameter adjustment results, environmental interference features, and behavioral feature data are fused in multiple dimensions to obtain the joint feature matrix.

[0086] It should be noted that historical access records are a complete log of all data access behaviors of the solid-state drive within a preset period (such as the past 24 hours or 7 days). Data access frequency refers to the total number of read and write requests accessing a specific data block per unit time, used to quantify the intensity of data access. The preset frequency threshold range needs to be set in conjunction with the actual use of the data block and the needs of the business scenario. For example, video data blocks used for real-time video transmission need to meet the low latency requirements for continuous viewing by users, so the frequency threshold is set to 120 times / minute; log data blocks used for system log storage only need to be backed up or audited during specific periods, so their frequency threshold is set to 10 times / minute. When the access frequency exceeds this threshold, it indicates that there is a system anomaly. The frequency value of the abnormal access, the time distribution of the anomaly, the duration, and the frequency fluctuation amplitude are recorded to form behavioral characteristic data.

[0087] In this embodiment, the current environmental fluctuation parameters refer to dynamic parameters in the real-time operating environment of the solid-state drive that may affect access efficiency, covering key indicators such as temperature, vibration frequency, and spatial position offset. Combined with behavioral characteristic data, a causal relationship between abnormal access behavior and environmental fluctuations can be established. For example, for every 1°C increase in temperature, data read latency increases by 2ms; for every 5 vibrations / second that exceed the normal range in vibration frequency, latency increases by 1ms. Based on these determined environmental interference characteristics, it is necessary to clarify the type of interference factor (temperature, vibration, etc.), its specific value, and its percentage impact on access efficiency.

[0088] In this embodiment, a preset weight threshold is obtained by collecting the average access latency of solid-state drives corresponding to different interference weight values. When the interference weight value is greater than the preset weight threshold, it indicates that the access latency caused by environmental interference exceeds the latency threshold of the corresponding data block, and environmental parameters need to be adjusted. The adjustment process requires the formulation of differentiated strategies based on the type of environmental fluctuation parameters and the purpose of the data block. For example, for video data blocks with latency caused by excessive temperature, the temperature of the area is precisely reduced to 25°C through the server room's zoned air conditioning system; for the problem of excessive vibration, the vibration frequency is controlled to within 45 times / second by optimizing the server fan speed. The obtained parameter adjustment results should include the adjusted target parameter values, adjustment methods, and expected optimization effects for different data blocks. Finally, the behavioral feature data, environmental interference features, and parameter adjustment results are structurally integrated according to the time dimension and feature dimension to obtain a joint feature matrix. The rows of the matrix represent preset time windows, consistent with the time interval in step S12, and the columns represent specific feature indicators, including behavioral dimension, environmental dimension, and adjustment dimension, which can capture the pattern of access patterns of data blocks with different purposes changing with the environment.

[0089] For example, a solid-state drive (SSD) in a server room stores two types of data blocks: video data blocks (ID: V001) and document data blocks (ID: D001). Pre-obtained 24-hour historical access records show that V001 is accessed 130 times / minute daily from 19:00 to 21:00, exceeding the preset frequency threshold of 120 times / minute, and is marked as abnormally high-frequency. D001's access frequency is stable at 40 times / minute, with its preset frequency threshold of 50 times / minute, showing no abnormality. Based on this, behavioral characteristic data is formed: V001's abnormal period is 19:00-21:00, with a frequency of 130 times / minute. Real-time environmental fluctuation parameters are a temperature of 30℃ and a vibration frequency of 55 vibrations / second. Combining the behavioral characteristic data, it is found that for every 1℃ that V001's temperature exceeds the normal range of 22-25℃, the latency increases by 0.6ms, and for every 1 vibration / second increase in vibration frequency, the latency increases by an additional 0.4ms. Therefore, the interference types are determined to be temperature and vibration, with temperature accounting for 60% and vibration for 40%. The interference weight value of V001 is calculated to be 0.6, which is greater than the preset weight threshold of 0.5. Parameter adjustment is initiated to adjust the temperature of the area where V001 is located to 25℃ and reduce the vibration frequency to 45 times / second.

[0090] Finally, the data is integrated into a joint feature matrix in 10-minute windows, including data block ID (V001), purpose (video transmission), access frequency (130 times / minute), anomaly marker (yes), current temperature (30℃), current vibration frequency (55 times / second), interference weight (0.6), adjusted temperature (25℃), and adjusted vibration frequency (45 times / second).

[0091] In step S14, based on the joint feature matrix, a long short-term memory network model is used to predict the distribution of hotspot data within a preset time period in the future, including:

[0092] Historical access frequency sequences, environmental parameter sequences, and delay data sequences of different types of data are extracted from the joint feature matrix and combined to construct a time series feature set.

[0093] The time series feature set is input into the pre-trained long short-term memory network model to obtain the temporal correlation features between access patterns and environmental factors;

[0094] Based on the aforementioned temporal correlation features, predict the trend of changes in the access frequency of different types of data within a future preset time period;

[0095] Based on the trend of access frequency changes, data blocks with access frequencies higher than a preset hotspot threshold are identified as the hotspot data distribution.

[0096] It should be noted that the historical access frequency sequence refers to the average access frequency change of different types of data blocks recorded within a preset period (usually 24 hours) according to a preset time window. For example, the sequence of average access frequencies per hour for video data blocks over the past 24 hours is [80, 85, ..., 150, 145], which can reflect the historical fluctuation pattern of data block access popularity. The environmental parameter sequence is a sequence formed by sorting environmental fluctuation data within the same time window by time, including temperature, vibration frequency, location offset, etc. The delay data sequence is a sequence of average access delay within the corresponding time window, sorted by time, which can reflect the change of access efficiency over time. These three types of sequences are combined after being aligned by time window to form a time series feature set.

[0097] It's worth noting that the Long Short-Term Memory (LSTM) network model is a deep learning model specifically designed for processing time series data. Its core advantage lies in its ability to effectively capture the dependencies between data over long time periods. The LSTM model consists of four layers: an input layer, hidden layers, fully connected layers, and an output layer. The input layer receives a pre-constructed time series feature set, with an input dimension of [T, ...]. F], where T is the length of the time series, i.e., the number of time windows within the preset period in the past, and F is the feature dimension of each time window, including four types of features: historical access frequency, temperature, vibration frequency, and delay. Therefore, F=4. The input data feature values ​​are normalized and linearly mapped to the interval [0,1]. The hidden layer contains two cascaded LSTM units, each with 128 neurons, including a forget gate, an input gate, and an output gate. The forget gate controls the retention ratio of historical information through the Sigmoid activation function. The input gate combines the Sigmoid and Tanh activation functions to filter and update the current features. The output gate controls the information transmission of cell state. The fully connected layer contains a ReLU activation function layer composed of 64 neurons and a single neuron transition layer, which compresses the temporal features into a 1-dimensional intermediate output. The output layer receives this 1-dimensional intermediate output and maps it to a dimension M that matches the number of future time windows through matrix transformation. For example, if the next 2 hours are divided into 1 hour / window, then M=2. The output value is then inversely normalized to obtain the actual predicted frequency.

[0098] In this embodiment, the LSTM model is trained using supervised learning. The training samples are derived from historical data from the past 30 days, split into training and validation sets in a 7:3 ratio. Mean squared error (MSE) is used as the loss function, the Adam optimizer is selected, the learning rate is set to 0.001, and the batch size is set to 8, achieving a balance between training speed and memory usage. The training epochs are set to 50. After each epoch, the MSE is calculated using the validation set. If the validation set MSE does not decrease for five consecutive epochs, a training stop mechanism is triggered. If convergence is not achieved after 50 epochs, the learning rate can be reduced to 0.0005 to continue training until the model converges. The trained model can learn cross-time correlation patterns such as "temperature increases by 2℃ at time t → video data access frequency increases by 10 times / minute at time t+1" and "vibration frequency exceeds 50 times / second at time t → document data access latency increases by 1ms at time t+30". The output temporal correlation features are the quantitative expression of these patterns.

[0099] In this embodiment, the preset time period needs to be set according to the scheduling requirements of the application scenario, with a common range of 1-6 hours. The prediction result will output the predicted access frequency values ​​of different types of data blocks within each time window. The obtained access frequency change trend refers to the dynamic evolution direction and magnitude of the access frequency of different types of data blocks over time within the preset time period. The hotspot threshold setting needs to be formulated according to the differences in access characteristics of data types. Specifically, due to the real-time transmission requirements, the hotspot threshold for video data blocks is set to 100 times / minute; the hotspot threshold for document data blocks is set to 80 times / minute; and the hotspot threshold for log data blocks is set to 10 times / minute. When the predicted access frequency value of a data block is consistently higher than its corresponding threshold, the data block is included in the hotspot data distribution, indicating that its storage resources need to be prioritized. If only a few windows exceed the threshold, such as document data blocks having a predicted frequency of 85 times / minute only from 20:00 to 20:10, while the frequency is lower than 80 times / minute in other windows, it is not listed as hotspot data to avoid misjudgment due to short-term fluctuations.

[0100] For example, the historical access frequency of video data block V001 in a server room shows a pattern of peak access between 19:00 and 21:00 daily (120-150 times / minute), with a gradual increase to 30-80 times / minute during other times. Historical data also shows that "for every 28°C increase in temperature, the access frequency one hour later is 10 times / minute higher than the average for the same period." The trained LSTM model has learned these two patterns. When predicting the access frequency between 18:00 and 19:00, the model automatically retains and utilizes the long-term pattern of "peak access at 19:00." By integrating the real-time environmental information of a current temperature of 30°C, the model ultimately predicts an access frequency of 130 times / minute between 19:00 and 20:00, calculated by adding the extra 10 times / minute due to temperature to the average of 120 times / minute. This prediction has an error of only 1.5% compared to the actual observed 128 times / minute, fully demonstrating the LSTM model's ability to capture long-term cyclical patterns and improving the accuracy of hotspot data distribution prediction.

[0101] In step S15, the current storage layout is obtained, and the matching degree between the hot data distribution and the current storage layout is calculated. If the matching degree is lower than a preset matching degree threshold, the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout, including:

[0102] Scan the storage metadata in the current storage layout, compare the storage metadata with the hot data distribution, calculate the proportion of hot data located in the high-latency area, and obtain the matching degree;

[0103] When the matching degree is lower than the preset matching degree threshold, high-frequency access data located in the high-latency region is selected from the hot data distribution and determined as the data set to be adjusted.

[0104] The data set to be adjusted is prioritized according to access frequency from high to low, and a migration target list is generated by combining the real-time storage capacity information of the low-latency area.

[0105] According to the migration target list, the high-frequency access data is migrated to the low-latency area in batches to obtain the storage layout after migration;

[0106] Calculate the matching degree between the migrated storage layout and the hot data distribution. If it is still lower than the preset matching degree threshold, adjust the data layering rules to obtain an optimized storage layout.

[0107] It's important to note that storage metadata reflects structured information about the storage status of all data blocks in a solid-state drive (SSD), specifically including key information such as the unique identifier of the data block, its physical storage address, the type of storage region it belongs to, and the capacity occupied by the data block. The type of storage unit in an SSD directly determines the basic access latency, which is the core physical basis for dividing low-latency and high-latency regions. Low-latency regions preferentially use SLC (Single-Level Cell) NAND flash memory, where each storage unit stores only 1 bit of data, resulting in a short electron migration path and fast read / write response. High-latency regions use MLC (Multi-Level Cell) or TLC (Triple-Level Cell) NAND flash memory, where each unit stores 2 or 3 bits of data, suitable for storing cold data with low access frequency and insensitivity to latency. The matching degree M is calculated by comparing the total amount of hot data (A) and the amount of hot data located in the high-latency region (B): M = 1 - B / A × 100%. A higher matching degree indicates that more hot data is located in the low-latency region, ensuring more efficient access.

[0108] The preset matching threshold is set based on the system's access latency requirements. When selecting the data set to be adjusted, it is necessary to re-verify whether the data access frequency is still higher than the preset hotspot threshold to avoid invalid migration due to a temporary drop in data popularity. Since data with higher access frequency is more sensitive to latency, prioritizing its migration can minimize overall access latency. Therefore, the migration target list should be sorted from highest to lowest access frequency. At the same time, it is necessary to ensure that the total amount of data to be migrated does not exceed the real-time remaining capacity of the low-latency zone to avoid capacity overflow and migration failure. Migrating a large amount of data at once will consume too much system bandwidth, causing queuing delays for normal data access requests at the same time. Batch migration can control the bandwidth consumption of each batch to within 30%, which will not affect normal data access services and can gradually complete the optimization, balancing migration efficiency and system stability. If the matching degree still does not meet the requirements after migration, the data tiering rules need to be adjusted, such as increasing the capacity ratio of the low-latency zone or optimizing the hotspot data identification logic, to further improve the matching degree and finally obtain an optimized storage layout that meets the latency requirements.

[0109] For example, a server room has a total solid-state drive capacity of 2TB, with a low-latency zone of 600GB and an access latency of 2ms, and a high-latency zone of 1400GB and an access latency of 12ms. The preset hotspot threshold is 100 times / minute. The current hotspot data distribution includes video data blocks V001 (20GB, 150 times / minute), V002 (30GB, 130 times / minute), and V003 (50GB, 110 times / minute), all of which are hotspot data. Scanning the storage metadata reveals that only V001 is located in the low-latency zone, while V002 and V003 are both located in the high-latency zone. The matching degree M is calculated as 1 - 80 / 100 × 100% = 20%, which is lower than the preset matching degree threshold of 80%. Therefore, V002 and V003 are included in the data set to be adjusted. The real-time remaining capacity of the low-latency zone is 580GB. The data to be adjusted is sorted by access frequency as [V002, V003], generating a migration target list. The first batch migrated 30GB of V002, the second batch migrated 30GB of V003, and the third batch migrated the remaining 20GB of V003, leaving 500GB of capacity in the low-latency zone. After the migration, the matching rate of hot data was 100%, and no adjustment of the tiering rules was required. The final optimized storage layout is as follows: V001, V002, and V003 are all located in the low-latency zone. The average latency for subsequent accesses can be reduced from the original 10ms to 2ms, significantly improving access efficiency.

[0110] In step S16, based on the optimized storage layout, the response time of data access requests is monitored in real time. When the response time exceeds a preset time threshold, log data containing latency feedback is generated, and latency load characteristics are extracted, including:

[0111] The response time of data access requests is continuously collected from the optimized storage layout;

[0112] When the response time exceeds a preset time threshold, log data containing latency feedback is generated, and access load fluctuation information is extracted from the log data.

[0113] When the fluctuation information exceeds a preset fluctuation threshold, the load change pattern in the fluctuation information is analyzed, and the delay load characteristics are obtained by combining the storage area information where the delay occurs.

[0114] It should be noted that the response time refers to the total time elapsed from initiating a data access request to the SSD returning the result. The preset time threshold needs to be set differently based on the data type and storage area characteristics. For video data blocks stored in the low-latency area, to meet the low-latency requirements of real-time transmission, the threshold is set to 3ms; for log data blocks stored in the high-latency area, which are less sensitive to latency, the threshold is set to 12ms. Log data includes the timestamp of the latency occurrence, the type of access request that triggered the latency (read request / write request), the corresponding data block ID and storage area, the actual response time, and concurrent environmental parameters, ensuring that subsequent analysis can trace the specific scenario in which the latency occurred.

[0115] In this embodiment, load fluctuation information refers to the change in the total number of access requests per unit time. Key quantitative indicators include instantaneous load (the number of access requests within a millisecond), average load (the average instantaneous load over a preset time period), and load fluctuation amplitude (represented by the standard deviation σ). The preset fluctuation threshold is set based on historical load stability data, typically taking the upper limit of the normal fluctuation range. For example, for video data blocks in low-latency areas, the historical load fluctuation standard deviation is mostly between 0.5 and 1 times / second, so the threshold is set to 1 time / second. Since different regions have different load-bearing capacities, when analyzing load change patterns, it is necessary to focus on the periodicity of load fluctuations, the specific time periods when fluctuation peaks occur, and the regional concentration characteristics of fluctuations. The final latency load characteristics include three core pieces of information: first, the standard deviation of load fluctuations; second, the correlation between load and latency; and third, the characteristics of the storage region corresponding to the fluctuations.

[0116] For example, in a cloud storage environment equipped with NVMe solid-state drives, the monitoring tool collected the response time of 1000 access requests per second. It was found that during the afternoon period of 14:00-14:30, the average response time of video data blocks in the low-latency zone reached 3.2ms, exceeding the preset time threshold of 3ms, and a latency log was subsequently generated. Access load fluctuation information was extracted from the log: the instantaneous load from 14:00-14:30 was 80-120 times / second, the average load μ=100 times / second, and the calculated load fluctuation standard deviation σ=0.9 times / second, exceeding the preset fluctuation threshold of 0.8 times / second. Further analysis of the load variation pattern revealed that the load peak was concentrated between 14:15 and 14:25, with fluctuations only occurring in the low-latency zone and a stable load in the high-latency zone. Ultimately, the latency load characteristics were extracted as follows: "The standard deviation of video data access load fluctuation in the low-latency zone was 0.9 times / second from 14:00 to 14:30 in the afternoon, and the load peak reached 120 times / second from 14:15 to 14:25. The response time increased linearly with the increase of load, and the load latency increased by 0.4ms for every 10 times / second increase."

[0117] In step S17, based on the latency load characteristics, the optimized storage layout is adjusted a second time by adjusting the data migration frequency and priority to obtain the final dynamic storage scheme, including:

[0118] Based on the latency load characteristics, high-frequency latency data and their corresponding latency anomaly regions are identified, and a list of data to be migrated is generated.

[0119] Obtain the remaining capacity of the current low-latency region, and combine it with the list of data to be migrated and the remaining capacity to determine the migration frequency table of the high-frequency latency data, thus obtaining the migration frequency table;

[0120] Based on the migration frequency table, the high-frequency latency data is prioritized and sorted from high to low access frequency to obtain a secondary migration target list.

[0121] Based on the secondary migration target list, the high-frequency latency data is migrated to the low-latency area in batches according to priority, resulting in the final dynamic storage scheme.

[0122] It should be noted that the high-frequency latency data refers to data blocks that simultaneously meet the criteria of an access frequency exceeding a preset hotspot threshold and a response time exceeding a preset time threshold. The latency anomaly region is the storage area where the high-frequency latency data is currently located, and its determination needs to be made in conjunction with the region load information in the latency load characteristics. If data latency occurs in a low-latency region due to overload, that low-latency region is a latency anomaly region; if high-frequency access data is mistakenly stored in a high-latency region, that high-latency region also belongs to the latency anomaly region. The migration frequency of high-frequency latency data is a quantitative plan for the migration rhythm, and its core components include migration batches, the time interval between each batch, and the amount of data migrated in each batch. If the remaining capacity in the current low-latency region is sufficient (e.g., 100GB of data to be migrated, 200GB remaining), the migration frequency can be appropriately increased to quickly resolve latency issues; if the remaining capacity is limited (e.g., 180GB of data to be migrated, 200GB remaining), the frequency needs to be reduced, and migration should be done in small batches to avoid a single migration occupying too much space and causing new capacity conflicts.

[0123] In this step, the secondary migration list still sorts high-frequency latency data from highest to lowest access frequency. It includes not only data migration priority but also the specific amount of data to be migrated, the corresponding access frequency, the target storage address for each batch, and the estimated migration completion time. For example, the first two items in a system's secondary migration list are: high-frequency latency data A (access frequency 250 times / minute, 60GB to be migrated, scheduled for the first batch from 10:00-10:20, target storage address is low-latency sector 0x5000-0x5AFF, estimated completion time 10:15) and high-frequency latency data B (access frequency 200 times / minute, 40GB to be migrated, scheduled for the second batch from 10:30-10:50, target storage address is low-latency sector 0x5B00-0x5FFF, estimated completion time 10:45). The resulting dynamic storage solution not only includes the adjusted storage layout but also ensures that when similar latency load characteristics occur again, the system can directly reuse or fine-tune parameters, achieving a dynamic optimization effect of "one-time adjustment, long-term adaptation".

[0124] In summary, this invention discloses an intelligent hierarchical storage method for hot and cold data in solid-state drives (SSDs). The method includes: collecting spatial location change data and vibration state data of the SSD during operation; preprocessing these data to obtain a multidimensional feature dataset; quantifying the efficiency impact of the spatial location change data and vibration state data based on the multidimensional feature dataset to obtain corresponding interference weight values; constructing a joint feature matrix based on the interference weight values ​​and pre-acquired historical access records; predicting the distribution of hot data within a preset time period using a long short-term memory network model based on the joint feature matrix; obtaining the current storage layout; calculating the matching degree between the hot data distribution and the current storage layout; if the matching degree is lower than a preset matching degree threshold, migrating high-frequency access data to a low-latency area to obtain an optimized storage layout; monitoring the response time of data access requests in real time based on the optimized storage layout; generating log data containing latency feedback and extracting latency load characteristics when the response time exceeds a preset time threshold; and further adjusting the optimized storage layout based on the latency load characteristics by adjusting the data migration frequency and priority to obtain the final dynamic storage scheme. This invention achieves dynamic optimization of solid-state drive storage performance and improves the overall system response speed by deeply integrating environmental interference and hot data with dynamic adjustment of storage layout.

[0125] Reference Figure 2 The second embodiment of the present invention provides a solid-state drive intelligent tiered storage system for hot and cold data, comprising:

[0126] The data acquisition and preprocessing module is used to collect spatial position change data and vibration state data of the solid-state drive during operation, perform preprocessing, and obtain a multi-dimensional feature dataset.

[0127] The interference weight quantization module is used to quantify the degree of efficiency impact of the spatial position change data and the vibration state data based on the multidimensional feature dataset, and obtain the corresponding interference weight value.

[0128] The feature matrix construction module is used to construct a joint feature matrix based on the interference weight values ​​and pre-acquired historical access records;

[0129] The hotspot data prediction module is used to predict the distribution of hotspot data within a preset time period based on the joint feature matrix and a long short-term memory network model.

[0130] The storage layout optimization module is used to obtain the current storage layout, calculate the matching degree between the hot data distribution and the current storage layout, and if the matching degree is lower than a preset matching degree threshold, then the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout.

[0131] The latency load extraction module is used to monitor the response time of data access requests in real time according to the optimized storage layout. When the response time is greater than a preset time threshold, log data containing latency feedback is generated and latency load characteristics are extracted.

[0132] The dynamic scheme adjustment module is used to make secondary adjustments to the optimized storage layout based on the latency load characteristics by adjusting the data migration frequency and priority, so as to obtain the final dynamic storage scheme.

[0133] It should be noted that the solid-state drive cold and hot data intelligent tiered storage system provided in this embodiment of the invention is used to execute all the process steps of the solid-state drive cold and hot data intelligent tiered storage method in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0134] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a hotspot data prediction program. When the processor executes the computer program, it implements the steps in the above embodiments of the intelligent tiered storage method for hot and cold data on solid-state drives, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the interference weight quantization module.

[0135] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0136] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0137] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0138] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0139] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0140] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0141] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent tiered storage of cold and hot data in a solid-state drive, characterized in that, include: The system collects spatial position change data and vibration status data of the solid-state drive during operation, performs preprocessing, and obtains a multidimensional feature dataset; among which, the spatial position change data refers to the real-time coordinate information of the solid-state drive in three-dimensional physical space. Based on the multidimensional feature dataset, the efficiency impact of the spatial location change data and the vibration state data is quantified to obtain the corresponding interference weight value. Based on the interference weight values ​​and the pre-acquired historical access records, a joint feature matrix is ​​constructed, and based on the joint feature matrix, a long short-term memory network model is used to predict the distribution of hot data within a future preset time period. The current storage layout is obtained, and the matching degree between the hot data distribution and the current storage layout is calculated. If the matching degree is lower than a preset matching degree threshold, the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout. Specifically, the storage metadata in the current storage layout is scanned, the storage metadata is compared with the hot data distribution, and the proportion of hot data located in the high-latency area is calculated to obtain the matching degree. Based on the optimized storage layout, the response time of data access requests is monitored in real time. When the response time is greater than a preset time threshold, log data containing latency feedback is generated, and latency load characteristics are extracted. The latency load characteristics include the standard deviation of load fluctuations, the correlation between load and latency, and the storage area characteristics corresponding to the fluctuations. In response to the aforementioned latency load characteristics, the optimized storage layout is further adjusted by adjusting the data migration frequency and priority to obtain the final dynamic storage solution. The step of constructing a joint feature matrix based on the interference weight value and pre-acquired historical access records includes: Data access frequency is extracted from the historical access records. If the data access frequency exceeds a preset frequency threshold, it is marked as abnormal behavior, and behavioral feature data is obtained. The current environmental fluctuation parameters are obtained, and the interference factors that cause the decrease in access efficiency are identified in combination with the behavioral feature data, and determined as environmental interference features. Based on the environmental interference features, when the interference weight value is greater than a preset weight threshold, the environmental fluctuation parameters are adaptively adjusted to obtain parameter adjustment results. The parameter adjustment results, the environmental interference features, and the behavioral feature data are fused in multiple dimensions to obtain the joint feature matrix.

2. The method for intelligent hierarchical storage of cold and hot data in a solid-state drive according to claim 1, characterized in that, The step of quantifying the efficiency impact of the spatial location change data and the vibration state data based on the multidimensional feature dataset to obtain corresponding interference weight values ​​includes: Based on the multidimensional feature dataset, the spatial coordinate offset corresponding to the spatial position change data is extracted, and time series analysis is performed on the vibration state data to obtain the corresponding vibration frequency data. The spatial coordinate offset is input into a pre-trained offset-interference regression model to obtain the position interference level, and the position interference factor is determined based on the position interference level. The vibration interference factor is obtained by matching the preset vibration interference mapping relationship with the vibration frequency data; The position interference factor and the vibration interference factor are nonlinearly combined to obtain a comprehensive interference weight value.

3. The intelligent tiered storage method for cold and hot data in a solid-state drive according to claim 1, characterized in that, The step of predicting the distribution of hotspot data within a preset time period using a long short-term memory network model based on the joint feature matrix includes: Historical access frequency sequences, environmental parameter sequences, and latency data sequences of different types of data are extracted from the joint feature matrix and combined to construct a time series feature set; wherein, the latency data sequence is a sequence of average access latency within the corresponding time window sorted by time; The time series feature set is input into the pre-trained long short-term memory network model to obtain the temporal correlation features between access patterns and environmental factors; Based on the aforementioned temporal correlation features, predict the trend of changes in the access frequency of different types of data within a future preset time period; Based on the trend of access frequency changes, data blocks with access frequencies higher than a preset hotspot threshold are included in the hotspot data distribution.

4. The intelligent tiered storage method for cold and hot data in a solid-state drive according to claim 3, characterized in that, If the matching degree is lower than a preset matching degree threshold, then high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout, including: When the matching degree is lower than the preset matching degree threshold, high-frequency access data located in the high-latency region is selected from the hot data distribution and determined as the data set to be adjusted. The data set to be adjusted is prioritized according to access frequency from high to low, and a migration target list is generated by combining the real-time storage capacity information of the low-latency area. According to the migration target list, the high-frequency access data is migrated to the low-latency area in batches to obtain the storage layout after migration; Calculate the matching degree between the migrated storage layout and the hot data distribution. If it is still lower than the preset matching degree threshold, adjust the data layering rules to obtain an optimized storage layout.

5. The intelligent tiered storage method for cold and hot data in a solid-state drive according to claim 1, characterized in that, The step involves monitoring the response time of data access requests in real time based on the optimized storage layout. When the response time exceeds a preset time threshold, log data containing latency feedback is generated, and latency load characteristics are extracted, including: The response time of data access requests is continuously collected from the optimized storage layout; When the response time exceeds a preset time threshold, log data containing latency feedback is generated, and access load fluctuation information is extracted from the log data. When the fluctuation information exceeds a preset fluctuation threshold, the load change pattern in the fluctuation information is analyzed, and the delay load characteristics are obtained by combining the storage area information where the delay occurs.

6. The intelligent tiered storage method for cold and hot data in a solid-state drive according to claim 1, characterized in that, The step of adjusting the optimized storage layout by adjusting the data migration frequency and priority in response to the latency load characteristics, to obtain the final dynamic storage solution, includes: Based on the latency load characteristics, high-frequency latency data and their corresponding latency anomaly regions are identified, and a list of data to be migrated is generated. Obtain the remaining capacity of the current low-latency region, and combine it with the list of data to be migrated and the remaining capacity to determine the migration frequency table of the high-frequency latency data, thus obtaining the migration frequency table; Based on the migration frequency table, the high-frequency latency data is prioritized and sorted from high to low access frequency to obtain a secondary migration target list. Based on the secondary migration target list, the high-frequency latency data is migrated to the low-latency area in batches according to priority, resulting in the final dynamic storage scheme.

7. A solid-state drive intelligent tiered storage system for hot and cold data, characterized in that, The method for implementing the intelligent tiered storage of hot and cold data in a solid-state drive as described in any one of claims 1 to 6 includes: The data acquisition and preprocessing module is used to collect spatial position change data and vibration state data of the solid-state drive during operation, perform preprocessing, and obtain a multi-dimensional feature dataset. The interference weight quantization module is used to quantify the degree of efficiency impact of the spatial position change data and the vibration state data based on the multidimensional feature dataset, and obtain the corresponding interference weight value. The feature matrix construction module is used to construct a joint feature matrix based on the interference weight values ​​and pre-acquired historical access records; The hotspot data prediction module is used to predict the distribution of hotspot data within a preset time period based on the joint feature matrix and a long short-term memory network model. The storage layout optimization module is used to obtain the current storage layout, calculate the matching degree between the hot data distribution and the current storage layout, and if the matching degree is lower than a preset matching degree threshold, then the high-frequency access data is migrated to a low-latency area to obtain an optimized storage layout. The latency load extraction module is used to monitor the response time of data access requests in real time according to the optimized storage layout. When the response time is greater than a preset time threshold, log data containing latency feedback is generated and latency load characteristics are extracted. The dynamic scheme adjustment module is used to make secondary adjustments to the optimized storage layout based on the latency load characteristics by adjusting the data migration frequency and priority, so as to obtain the final dynamic storage scheme.

8. An electronic device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the solid-state drive intelligent tiered storage method for hot and cold data as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the intelligent tiered storage method for cold and hot data of a solid-state drive as described in any one of claims 1 to 6.

Images