A solid-state hard drive read and write rate analysis method and system

Abstract

The present invention provides a solid-state drive (SSD) read / write rate analysis method and system, relating to the field of SSD technology. The method comprises: collecting SSD status information and read / write rate; analyzing influencing indicators of the SSD read / write rate using the extension analytic hierarchy process (AHP); and making a preliminary prediction of the SSD read / write rate using a fuzzy proximity calculation method; co-simulating the SSD based on digital twin technology; and constructing a SSD read / write rate compensation model in combination with a recurrent neural network; and using the read / write rate compensation model to perform error correction on the preliminary prediction results of the SSD read / write rate to obtain the final SSD read / write rate. The present invention not only improves the prediction accuracy of SSD read / write rates but also provides strong support for SSD performance evaluation, system optimization, and hardware selection.

Description

Technical Field

[0001] The present invention relates to the technical field of solid-state hard disks, and more particularly to a method and system for analyzing the read and write speeds of solid-state hard disks. Background Art

[0002] The rapid development and widespread adoption of computing technologies such as the internet, cloud computing, and the Internet of Things (IoT) have generated unprecedented amounts of data in our daily lives and work, requiring efficient processing and storage solutions. This explosive growth, users' increasing focus on data security, and the continuous advancement of information technology are placing ever-demanding demands on storage system performance.

[0003] Solid-state drives (SSDs), a new type of hard drive based on an array of solid-state electronic memory chips, have become a leader in the storage field thanks to their structure consisting of a control unit and solid-state storage units (such as NAND flash memory chips). The stable nature of flash memory, particularly the widespread use of NAND flash memory chips, has made them a key indicator of the quality of SSDs. Read and write performance is a key metric for SSDs, directly impacting data processing efficiency and user experience.

[0004] However, the read and write speeds of SSDs are not static but are influenced by a combination of factors. Factors such as temperature, load, total bytes written (TBW), and drive health all influence SSD performance in their own unique ways. These factors are not only diverse in their dimensions but also intricately interrelated, causing SSD behavior to vary significantly under different operating conditions. Traditional methods for predicting drive read and write speeds, such as linear regression or simple statistical models, often struggle to capture this complexity. They ignore the dynamic changes in drives under different environments, resulting in inaccurate predictions in real-world applications.

[0005] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0006] In view of this, the present invention provides a solid state drive read and write rate analysis method and system to solve the above-mentioned problems.

[0007] In order to solve the above problems, the specific technical solutions adopted by the present invention are as follows:

[0008] According to one aspect of the present invention, a method for analyzing the read and write speeds of a solid-state drive is provided, comprising:

[0009] S1. Collect the status information and read / write speed of the solid-state drive, use the extension analytic hierarchy process to analyze the influencing indicators of the solid-state drive read / write speed, and make a preliminary prediction of the solid-state drive read / write speed through the fuzzy proximity calculation method;

[0010] S2. Based on digital twin technology, we conduct joint simulation of solid-state drives and build a read and write rate compensation model for solid-state drives using a recurrent neural network.

[0011] S3. Use the read / write rate compensation model to perform error correction on the preliminary prediction result of the solid-state drive read / write rate to obtain the final solid-state drive read / write rate.

[0012] Preferably, the collecting of the state information and read / write speed of the solid-state hard disk, analyzing the influencing indicators of the read / write speed of the solid-state hard disk using the extension analytic hierarchy process, and making a preliminary prediction of the read / write speed of the solid-state hard disk using the fuzzy proximity calculation method includes:

[0013] S11, using the state information of the solid-state drive as different influencing indicators that affect the read and write speed of the solid-state drive, and constructing an extension interval number judgment matrix by comparing the indicators pairwise;

[0014] S12. Based on the extension interval number judgment matrix, the extension hierarchical analysis method is used to calculate the initial weight of each influencing indicator, and the initial weight is corrected by the order relationship analysis method to obtain the final indicator weight;

[0015] S13. Based on the read and write speed of the solid-state drive, data fitting technology is used to determine the fitting relationship between each influencing indicator and the read and write speed of the solid-state drive. Combined with the final indicator weight, the fuzzy proximity calculation method is used to make a preliminary prediction of the read and write speed of the solid-state drive.

[0016] Preferably, the initial weights of the influencing indicators are calculated based on the extension interval number judgment matrix using the extension hierarchical analysis method, and the initial weights are corrected by the order relationship analysis method to obtain the final indicator weights, which include:

[0017] S121, performing a consistency index test on the extension interval number judgment matrix, and calculating the normalized eigenvector corresponding to the maximum eigenroot of the extension interval number judgment matrix that passes the consistency test to obtain a normalized weight vector;

[0018] S122. Form a weight interval vector that satisfies the consistency condition based on the normalized weight vector, and calculate the single ranking weight of each influencing indicator corresponding to its upper-level indicator based on the weight interval vector;

[0019] S123, calculating the group decision average indicator weight vector based on the single ranking weight to obtain the initial weight;

[0020] S124. Use the ordinal relationship analysis method to determine the importance ranking of the subjective weights and calculate the relative importance ratio. According to the importance ranking of the subjective weights and the relative importance ratio, modify the initial weights to obtain the final indicator weights.

[0021] Preferably, the method of determining the fitting relationship between each influencing indicator and the read / write rate of the solid-state drive using a data fitting technique based on the read / write rate of the solid-state drive, and performing a preliminary prediction of the read / write rate of the solid-state drive using a fuzzy proximity calculation method in combination with the final indicator weight includes:

[0022] S131, preprocessing the read and write rates of the solid-state drive, wherein the preprocessing includes: removing noise, filling missing values, and normalizing;

[0023] S132, fitting each influencing indicator and the read / write rate using nonlinear fitting to obtain a fitting relationship between each indicator and the read / write rate of the solid-state drive;

[0024] S133. Establish training samples based on the fitting relationship between each indicator and the read and write speed of the solid-state drive. Calculate the closeness between the training samples and the prediction samples based on the final indicator weights, and make a preliminary prediction of the read and write speed using the weighted average method.

[0025] Preferably, the steps of establishing a training sample based on the fitting relationship between each indicator and the read / write rate of the solid-state drive, calculating the closeness between the training sample and the prediction sample in combination with the final indicator weight, and making a preliminary prediction of the read / write rate by a weighted average method include:

[0026] S1331, using the pre-processed read / write rate as the sample output and the read / write rate influencing index as the sample input to form a training sample matrix;

[0027] S1332. Obtain prediction samples of the read / write rates of the solid-state drive to be predicted, and calculate the closeness between the training samples and the prediction samples using a closeness calculation formula based on the final indicator weights;

[0028] S1333. Sort the obtained closeness, select a preset number of training samples according to the sorting result, and predict the read and write rates of the solid-state drive to be tested by a weighted average method to obtain a preliminary prediction result.

[0029] Preferably, the closeness calculation formula is:

[0030] ;

[0031] ;

[0032] ;

[0033] Where, H ( T , T ′) represents the closeness, Ti Indicates the i training samples, T ′ represents the sample to be predicted, T i ⊕ T ′ represents the union operation of fuzzy sets, It is represented as the intersection operation of fuzzy sets, Y represents the set of read and write rates corresponding to all influencing indicators, y ki Indicates the i Of the training samples k The reading and writing rates corresponding to the impact indicators are: y′ k Indicates the first k The reading and writing rates corresponding to the impact indicators are: w k Indicates the k The final indicator weight of each influencing indicator.

[0034] Preferably, the method of co-simulating the solid-state drive based on the digital twin technology and building a read / write rate compensation model for the solid-state drive in combination with a recurrent neural network includes:

[0035] S21. Use digital twin technology to build a virtual model of the solid-state drive, and use the virtual model of the solid-state drive to perform joint simulations under different conditions to obtain simulated read and write rate data;

[0036] S22. Combining the theoretical read and write rate data and simulated read and write rate data of solid-state drives, a data-driven read and write rate compensation model is constructed based on a convolutional neural network integrated with an attention mechanism.

[0037] Preferably, the method of combining theoretical read / write rate data and simulated read / write rate data of the solid-state drive and building a data-driven read / write rate compensation model based on a convolutional neural network fused with an attention mechanism includes:

[0038] S221. Determine theoretical read / write rate data of the solid-state drive under the same conditions based on the simulated read / write rate data of the solid-state drive, and calculate the difference between the simulated read / write rate data and the theoretical read / write rate data to obtain a read / write rate error.

[0039] S222. Using a convolutional neural network, extract local features using the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data, and perform weighted processing on the extracted local features using a channel attention mechanism to generate a weighted feature map.

[0040] S223. Based on the weighted feature map, a data-driven read / write rate compensation model is constructed through a fully connected layer.

[0041] Preferably, the method of extracting local features by using a convolutional neural network and utilizing the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data, and performing weighted processing on the extracted local features by utilizing a channel attention mechanism to generate a weighted feature map includes:

[0042] S2221, inputting the read / write rate error into a convolutional neural network, and extracting local features from the input read / write rate error data through a convolution operation;

[0043] S2222, performing nonlinear mapping on the local features after the convolution operation through an activation function to obtain a feature map after nonlinear mapping;

[0044] S2223. Use the channel attention mechanism to perform weighted processing on the feature map after nonlinear mapping to obtain a weighted feature map.

[0045] According to another aspect of the present invention, a solid-state hard disk read and write rate analysis system is provided, comprising:

[0046] The preliminary prediction module is used to collect the status information and read and write speed of the solid-state drive, analyze the influencing indicators of the solid-state drive read and write speed using the extension hierarchical analysis method, and make a preliminary prediction of the solid-state drive read and write speed through the fuzzy proximity calculation method;

[0047] A compensation model building module is used to co-simulate solid-state drives based on digital twin technology and build a read and write rate compensation model for solid-state drives using a recurrent neural network.

[0048] The read / write rate correction module is used to use the read / write rate compensation model to perform error correction on the preliminary prediction results of the solid-state hard disk read / write rate to obtain the final solid-state hard disk read / write rate.

[0049] The beneficial effects of the present invention are:

[0050] 1. This invention combines multiple advanced technologies and methods, demonstrating strong adaptability and scalability. As SSD technology continues to evolve and the hardware environment changes, the accuracy and effectiveness of this method can be maintained by updating the digital twin model and adjusting recurrent neural network parameters. This not only improves the accuracy of SSD read and write rate predictions but also provides strong support for SSD performance evaluation, system optimization, and hardware selection.

[0051] 2. The present invention uses the extension hierarchical analysis method to calculate the initial weight of each influencing indicator, and corrects the initial weight through the ordinal relationship analysis method to obtain a more reasonable final indicator weight. The data fitting technology is used to determine the fitting relationship between each influencing indicator and the read and write rate of the solid-state hard disk, which can accurately reflect the intrinsic connection between the indicator and the read and write rate. The fuzzy proximity calculation method is used to make a preliminary prediction of the read and write rate of the solid-state hard disk, taking into account the fuzziness and similarity between samples. By calculating the proximity between the training sample and the predicted sample, and selecting a preset number of training samples for weighted average prediction, the prediction result is closer to the actual situation.

[0052] 3. The present invention uses digital twin technology to construct a virtual model of the solid-state drive, which can realize real-time simulation and emulation of the state of the solid-state drive. By integrating the attention mechanism, the convolutional neural network can pay more attention to the features that have a greater impact on the read and write rates, thereby improving the sensitivity and accuracy of the model. The introduction of the attention mechanism enables the model to dynamically adjust the attention to different features, thereby enhancing the adaptability and robustness of the model. By calculating the difference between the simulated read and write rate data and the theoretical read and write rate data, the read and write rate error is obtained as the input of the compensation model. The data-driven compensation model can make full use of existing data and improve the accuracy and reliability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0054] Figure 1 is a flow chart of a method for analyzing read and write rates of a solid-state hard disk according to an embodiment of the present invention;

[0055] Figure 2 This is a principle block diagram of a solid-state hard disk read and write rate analysis system according to an embodiment of the present invention.

[0056] In the picture:

[0057] 1. Preliminary prediction module; 2. Compensation model construction module; 3. Read and write rate correction module. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0059] According to an embodiment of the present invention, a method and system for analyzing the read and write rates of a solid-state drive are provided.

[0060] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a method for analyzing the read and write rates of a solid-state hard disk is provided, comprising:

[0061] S1. Collect the status information and read / write speed of the solid-state drive, use the extension analytic hierarchy process to analyze the influencing indicators of the solid-state drive read / write speed, and make a preliminary prediction of the solid-state drive read / write speed through the fuzzy proximity calculation method;

[0062] It should be noted that the status information of a hard drive includes many factors that affect its read and write speeds. Common influencing indicators include:

[0063] Temperature: The temperature of an SSD generally affects its performance; excessively high temperatures can cause it to slow down.

[0064] Remaining available space: The remaining space of the solid-state drive affects the read and write speeds. The less space there is, the lower the write speed may be.

[0065] Block erase count (Total Bytes Written): This indicator reflects the degree of wear of the solid-state drive. The more severe the wear, the more likely the performance will decline.

[0066] Drive health: An SSD with poor health may exhibit lower read and write rates.

[0067] Input / output operations per second (IOPS): An important indicator reflecting the hard disk's response performance, affecting data transmission speed.

[0068] Power supply status: Voltage fluctuations or unstable power supply may cause hard drive performance to degrade.

[0069] As a preferred embodiment, the collecting of the state information and read / write speed of the solid-state drive, analyzing the influencing indicators of the read / write speed of the solid-state drive using the extension analytic hierarchy process, and making a preliminary prediction of the read / write speed of the solid-state drive using the fuzzy proximity calculation method include:

[0070] S11, using the state information of the solid-state drive as different influencing indicators that affect the read and write speed of the solid-state drive, and constructing an extension interval number judgment matrix by comparing the indicators pairwise;

[0071] It should be noted that, through the experience of experts or existing data, these influencing indicators are compared pairwise to construct the extension interval number judgment matrix.

[0072] Extension interval numbers are a method for describing and handling uncertainty, particularly in complex decision-making problems. They can describe the relationships between pairs of indicators when comparing them, while also accounting for the uncertainty of these relationships.

[0073] Experts or relevant data analysts will assign relative importance scores to each pair of influencing indicators based on their importance, using a pairwise comparison method. Scoring can be done using a 1 to 9 scale (AHP scale) or other similar methods. For each pair of influencing indicators, experts need to answer the following questions:

[0074] index A Relative to the indicator B How important is it?

[0075] For example, if A Relative to B If it is "extremely important", it may be given a very high score (such as 9); if it is "somewhat important", it may be given a low score (such as 3 or 4).

[0076] Introduction of Extension Interval Numbers: Uncertainty often arises in pairwise comparisons (for example, a metric may have a higher priority in some contexts and a lower priority in others). To address this uncertainty, extension interval numbers are employed. For each pair of metric comparisons, the constructed judgment matrix is not simply a single value, but rather an interval that represents the range of metric importance. For example, the importance between metrics can be represented using intervals, such as [a, b], where a and b represent the minimum and maximum values for the metric pair, respectively. This accounts for the ambiguity of subjective judgments.

[0077] Assume there are three impact indicators T 1 (temperature), T 2 (remaining space), and T 3 (total number of bytes written). After pairwise comparison, the resulting judgment matrix is as follows:

[0078] ;

[0079] Among them: "1" in the first row and first column means T 1 andT The importance ratio of 1 itself is equal.

[0080] The interval value "[2,3]" in the first row and second column indicates that the experts believe T 1 to T The importance level of 2 is between "2 to 3".

[0081] The interval value "[3,4]" in the first row and third column indicates that the experts believe T 1 to T The importance level of 3 is between "3 and 4".

[0082] S12. Based on the extension interval number judgment matrix, the extension hierarchical analysis method is used to calculate the initial weight of each influencing indicator, and the initial weight is corrected by the order relationship analysis method to obtain the final indicator weight;

[0083] As a preferred embodiment, the initial weights of the influencing indicators are calculated based on the extension interval number judgment matrix using the extension hierarchical analysis method, and the initial weights are corrected by the order relationship analysis method to obtain the final indicator weights including:

[0084] S121, performing a consistency index test on the extension interval number judgment matrix, and calculating the normalized eigenvector corresponding to the maximum eigenroot of the extension interval number judgment matrix that passes the consistency test to obtain a normalized weight vector;

[0085] Specifically, the purpose of consistency test is to ensure that the judgment result of the extension interval number judgment matrix is reasonable, that is, the relative importance ratios of the indicators in the matrix are consistent. In the traditional hierarchical analysis method, this test is performed by calculating the consistency ratio ( CR For the extension interval number judgment matrix, a similar method can also be used to perform consistency check.

[0086] Calculate the consistency index CI , and through the maximum characteristic root λ max Calculating the consistency ratio CR .

[0087] For consistency testing of the matrix of extension interval numbers, a fuzzy consistency test method can be used to consider the fuzziness and uncertainty of the interval numbers. If the test passes, the matrix can be used for subsequent eigenvector calculations.

[0088] By solving the characteristic equation of the matrix of extension interval number judgment, we can get its maximum characteristic root λ max and the corresponding eigenvectors.

[0089] The step of normalizing the feature vector is to divide each component of the feature vector by the sum of the vector to ensure that all weights add up to 1. The resulting normalized feature vector represents the preliminary weight of each impact indicator.

[0090] In addition, the biggest feature λ max and consistency index CI Calculation formula:

[0091] ;

[0092] in, n Indicates the order of the judgment matrix (the number of influencing indicators).

[0093] S122. Form a weight interval vector that satisfies the consistency condition based on the normalized weight vector, and calculate the single ranking weight of each influencing indicator corresponding to its upper-level indicator based on the weight interval vector;

[0094] It's important to note that the weight interval vector is formed by expanding the normalized weight vector to form the weight interval for each influencing indicator, thereby representing uncertainty and ambiguity. Specifically, the weight interval vector expresses the range of each indicator's weight by assigning an upper and lower bound (minimum and maximum values) to the weight of that indicator.

[0095] The single ranking weight is to calculate the relative importance of each influencing indicator relative to its upper-level indicator based on the weight interval of each influencing indicator. In the hierarchical analysis method, the weight of each lower-level indicator is usually determined based on the weight of its upper-level indicator. The specific steps include:

[0096] Determine the parent metric: Assuming there are multiple parent metrics, the weight interval vector helps determine the relative importance of each child metric relative to the parent metric. For example, child metrics that affect SSD read and write speeds might be "temperature," "free space," and "total bytes written." Their ranking weights relative to the parent metric, "drive performance," need to be calculated.

[0097] Weighting of interval weights: Based on the weight interval vector of each indicator, the single ranking weight of the lower-level indicators can be calculated by interval weighting. For example, assuming that the weights of the upper-level indicators of "temperature" and "remaining space" are w 1=[0.45,0.55] and w 2=[0.25,0.35], then the single ranking weights of the subordinate indicators can be calculated based on these intervals.

[0098] Calculating Single Ranking Weights: Single ranking weights are typically calculated by normalizing and weighting the weight interval vectors. During pairwise comparisons, the resulting weight interval vectors influence the importance ranking of the lower-level indicators relative to the upper-level indicators. For a given weight interval, the single ranking weights of the lower-level indicators can be calculated using a weighted approach based on the number of intervals.

[0099] S123, calculating the group decision average indicator weight vector based on the single ranking weight to obtain the initial weight;

[0100] It should be noted that the group decision method obtains the final decision result by taking a weighted average of the judgment results of multiple experts. In this step, the group decision average weight vector represents the comprehensive result of all expert opinions.

[0101] The average weight of group decision making for each influencing indicator is calculated by usually taking the weighted average of the judgments of each expert on the indicator (such as the single ranking weight).

[0102] The average weight vector of group decision making is the initial weight of all influencing indicators. These weights reflect the relative importance of each influencing indicator in the overall decision.

[0103] S124. Use the ordinal relationship analysis method to determine the importance ranking of the subjective weights and calculate the relative importance ratio. According to the importance ranking of the subjective weights and the relative importance ratio, modify the initial weights to obtain the final indicator weights.

[0104] It's important to note that ranking analysis is a technique that helps decision makers rank indicators by comparing their relative importance. This method effectively determines the importance ranking of each indicator and calculates the relative importance ratios between indicators. This is particularly useful for revising initial weights derived from multiple experts or different data sources. In particular, when weights are inconsistent, it allows for integrating different perspectives and optimizing the final indicator weights.

[0105] In the ordinal relationship analysis method, the first step is to determine the importance of each indicator relative to other indicators based on subjective judgment or expert opinion. Usually, experts will rank the indicators based on experience or historical data and give the relative importance of each indicator. For example, suppose there are three indicators T 1. T 2 and T 3. Experts rank them based on their impact on the read and write speed of the solid-state drive. For example:

[0106] T 1 (temperature)> T 2 (remaining space)> T3 (total bytes written);

[0107] This means that experts believe that the temperature ( T 1) The biggest impact on the read and write speed of the solid-state drive, followed by the remaining space ( T 2), and finally the total number of bytes written ( T 3).

[0108] In addition, after determining the ranking of subjective weights, the next step is to calculate the relative importance ratios between the indicators. The relative importance ratio describes the importance of one indicator relative to another. In the ordinal relationship analysis method, the following method is usually used for calculation:

[0109] if T 1 is more important than T 2, then set the ratio r 12 >1, and the specific value can be determined based on expert opinion.

[0110] if T 1 and T 2 are of equal importance, then set the ratio r 12 =1.

[0111] if T 1 is less important than T 2, then set the ratio r 12 <1.

[0112] Then the relative importance ratio examples are:

[0113] Assumed temperature T 1 relative to the remaining space T The ratio of 2 is 2:1 ( r 12 =2), indicating T The importance of 1 is T Twice as much as 2.

[0114] Assuming the remaining space T 2 relative to the total number of bytes written T The ratio of 3 is 3:1 (i.e. r 23 =3).

[0115] In addition, once the importance ranking and relative importance ratio of subjective weights are obtained, these ratios can be used to modify the initial weights. The key to the modification process is to redistribute the weight of each indicator based on the ranking and ratio. The following is the modification method:

[0116] Adjust weight ratio: By calculating the relative importance ratio, the initial weight can be modified. For example, if the initial weight vector is W init =[ w 1, w 2, w 3], and the ratio matrix is R , the weights can be adjusted according to the ratio matrix.

[0117] Weighted average method: Use the weighted average method to modify the initial weight. The weighted average method is based on the relative importance ratio. r ij To determine the final weight of each indicator. For example, assuming the initial weight is W init =[0.5,0.3,0.2], the relative importance ratio matrix is R , the final weight of each indicator can be calculated by weighted average.

[0118] Through weighted correction using the ordinal relationship analysis method, the final weight of each influencing indicator can be obtained. These weights can more accurately reflect the actual importance of each indicator relative to the read and write speed of the solid-state drive.

[0119] For example: Assume that the weights corrected by the ordinal relationship analysis method are:

[0120] W final =[0.55,0.35,0.10];

[0121] Then it means: Temperature ( T 1) has a final weight of 0.55; the remaining space ( T 2) has a final weight of 0.35; the total number of bytes written ( T The final weight of 3) is 0.10.

[0122] S13. Based on the read and write speed of the solid-state drive, data fitting technology is used to determine the fitting relationship between each influencing indicator and the read and write speed of the solid-state drive. Combined with the final indicator weight, the fuzzy proximity calculation method is used to make a preliminary prediction of the read and write speed of the solid-state drive.

[0123] As a preferred embodiment, based on the read and write speed of the solid-state drive, the data fitting technology is used to determine the fitting relationship between each influencing indicator and the read and write speed of the solid-state drive, and the fuzzy proximity calculation method is used to perform a preliminary prediction of the read and write speed of the solid-state drive in combination with the final indicator weight, including:

[0124] S131, preprocessing the read and write rates of the solid-state drive, wherein the preprocessing includes: removing noise, filling missing values, and normalizing;

[0125] It should be noted that the purpose of preprocessing is to ensure data quality and usability. Read and write rate data is often affected by noise, missing values, or outliers, so it needs to be cleaned and processed.

[0126] Common preprocessing steps include:

[0127] Denoising: Remove abnormal values or outliers from read and write rate data.

[0128] Missing value filling: Missing read and write rate data can be filled through interpolation, mean filling, or other methods.

[0129] Normalization: Normalize the read and write rate data to make it uniform in range, usually standardizing the data to between 0 and 1.

[0130] S132, fitting each influencing indicator and the read / write rate using nonlinear fitting to obtain a fitting relationship between each indicator and the read / write rate of the solid-state drive;

[0131] Specifically, common nonlinear fitting methods include:

[0132] Quadratic polynomial fitting: Use a polynomial function to fit the relationship between the impact indicator and the read and write rate.

[0133] Support Vector Regression (SVR): Nonlinear fitting is performed using the regression algorithm of the support vector machine.

[0134] Neural Network Fitting: Use neural network models in deep learning to learn complex nonlinear relationships.

[0135] For example, suppose a quadratic polynomial model is selected to fit the relationship between read and write rates and impact indicators. After fitting, the following model is obtained:

[0136] R =1.2+0.5 T 1+0.3 T 2+0.1 T 3+0.02 T 1 2 +0.05 T 2 2 +0.01 T 3 2 ;

[0137] in, a 0=1.2, a 1=0.5, a 2=0.3, a 3=0.1,… are the coefficients obtained by fitting.

[0138] S133. Establish training samples based on the fitting relationship between each indicator and the read and write speed of the solid-state drive. Calculate the closeness between the training samples and the prediction samples based on the final indicator weights, and make a preliminary prediction of the read and write speed using the weighted average method.

[0139] As a preferred embodiment, the method of establishing training samples based on the fitting relationship between each indicator and the read and write speed of the solid-state drive, calculating the closeness between the training samples and the prediction samples in combination with the final indicator weight, and making a preliminary prediction of the read and write speed by the weighted average method includes:

[0140] S1331, using the pre-processed read / write rate as the sample output and the read / write rate influencing index as the sample input to form a training sample matrix;

[0141] In addition, various indicators that affect the read and write speed of the SSD (such as temperature, free space, and total bytes written) are combined with the preprocessed read and write speed data to form a training sample matrix. Each row represents a sample, listing the corresponding influencing indicator and its corresponding read and write speed.

[0142] Assume there are 3 influencing indicators: temperature T 1. Remaining space T 2 and the total number of bytes written T 3, and the corresponding read and write rates R The sample matrix X can be expressed as:

[0143] ;

[0144] in, T 1 (i) , T 2 (i) , T 3 (i) Indicates the i The impact index value of each sample, R (i) Indicates the corresponding read and write rate.

[0145] S1332. Obtain prediction samples of the read / write rates of the solid-state drive to be predicted, and calculate the closeness between the training samples and the prediction samples using a closeness calculation formula based on the final indicator weights;

[0146] As a preferred implementation, the closeness calculation formula is:

[0147] ;

[0148] ;

[0149] ;

[0150] Where, H ( T , T ′) represents the closeness, T i Indicates the i training samples, T ′ represents the sample to be predicted, T i ⊕ T ′ represents the union operation of fuzzy sets, It is represented as the intersection operation of fuzzy sets, Y represents the set of read and write rates corresponding to all influencing indicators, y ki Indicates the i Of the training samples k The reading and writing rates corresponding to the impact indicators are: y′ k Indicates the first k The reading and writing rates corresponding to the impact indicators are: w k Indicates the k The final indicator weight of each influencing indicator.

[0151] S1333. Sort the obtained closeness, select a preset number of training samples according to the sorting result, and predict the read and write rates of the solid-state drive to be tested by a weighted average method to obtain a preliminary prediction result.

[0152] Specifically, the training samples are ranked based on the calculated closeness between them and the samples to be predicted. The training samples with greater closeness are considered more representative and should have a greater weight in the final prediction. m The more training samples selected, the more accurate the prediction results will be.

[0153] For the selected training samples, the read and write rates are predicted using a weighted average method. The weighted average method combines the closeness of each training sample with the read and write rates to calculate the read and write rates of the SSD to be predicted.

[0154] The weighted average method formula is:

[0155] ;

[0156] Where, R predict Represents the preliminary predicted read and write rate, R i Indicates the i The read and write rate of training samples.

[0157] S2. Based on digital twin technology, we conduct joint simulation of solid-state drives and build a read and write rate compensation model for solid-state drives using a recurrent neural network.

[0158] As a preferred embodiment, the method of co-simulating the solid-state drive based on digital twin technology and building a read and write rate compensation model for the solid-state drive in combination with a recurrent neural network includes:

[0159] S21. Use digital twin technology to build a virtual model of the solid-state drive, and use the virtual model of the solid-state drive to perform joint simulations under different conditions to obtain simulated read and write rate data;

[0160] It's important to note that word twinning technology simulates the physical and behavioral characteristics of a solid-state drive (SSD) by creating a virtual model. This virtual model not only replicates the drive's structure and operating state, but also simulates operating conditions in different environments, including the impact of variables like temperature, humidity, TBW (total bytes written), and free space on read and write speeds.

[0161] A digital twin model of the solid-state drive is constructed using advanced simulation tools (such as ANSYS and COMSOL). This model is based on the drive's physical characteristics, operating mechanisms, and external environmental factors.

[0162] Based on the digital twin model, simulations are conducted under various conditions to simulate the SSD's performance under different working environments and operating conditions. Simulations can generate different read and write rate data by adjusting environmental factors (such as temperature, load, and drive health). Through co-simulation, SSD read and write rate data under different environmental and operating conditions can be obtained. This data will be used for subsequent comparisons between theoretical and simulated data to calculate errors and provide training data for compensation models.

[0163] S22. Combining the theoretical read and write rate data and simulated read and write rate data of solid-state drives, a data-driven read and write rate compensation model is constructed based on a convolutional neural network integrated with an attention mechanism.

[0164] As a preferred embodiment, the method of combining theoretical read and write rate data and simulated read and write rate data of solid-state drives and building a data-driven read and write rate compensation model based on a convolutional neural network integrated with an attention mechanism includes:

[0165] S221. Determine theoretical read / write rate data of the solid-state drive under the same conditions based on the simulated read / write rate data of the solid-state drive, and calculate the difference between the simulated read / write rate data and the theoretical read / write rate data to obtain a read / write rate error.

[0166] It should be noted that theoretical read and write speed data is based on the SSD's design parameters, technical specifications, and test results under ideal conditions. Typically, this data can be obtained from the SSD manufacturer's technical specifications or standard test environment.

[0167] Technical Specifications: The technical documentation provided by the hard drive manufacturer usually lists the maximum theoretical read and write speeds of the solid-state drive. For example, suppose a certain solid-state drive has a theoretical maximum read speed of 550MB / s and a write speed of 500MB / s.

[0168] Standard test conditions: Theoretical read and write speeds are generally measured under specific standard conditions, such as a temperature of 25°C, a humidity of 40%, and optimal hard drive health.

[0169] For example, under ideal conditions, the theoretical read rate is 550MB / s and the theoretical write rate is 500MB / s.

[0170] In addition, to measure the difference between simulated data and theoretical data, the read and write rate error can be calculated, usually using absolute error or relative error.

[0171] S222. Using a convolutional neural network, extract local features using the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data, and perform weighted processing on the extracted local features using a channel attention mechanism to generate a weighted feature map.

[0172] As a preferred embodiment, the method of extracting local features by using the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data through a convolutional neural network, and weighting the extracted local features using a channel attention mechanism to generate a weighted feature map includes:

[0173] S2221, inputting the read / write rate error into a convolutional neural network, and extracting local features from the input read / write rate error data through a convolution operation;

[0174] It's important to note that a convolutional neural network performs a convolution operation on this error data. This operation scans the input data through different convolution kernels (filters) to extract local features from the input data. Each convolution kernel detects a specific local pattern or trend.

[0175] Convolution operation formula:

[0176] Assume the input data is P , the convolution kernel is W , the output feature map is Q , the convolution operation can be expressed as:

[0177] ；

[0178] in, represents the convolution operation, W is the convolution kernel, P is the input data.

[0179] Through the convolution operation, the model can extract local features from the error data, which may be the pattern of read / write rate errors under different environmental conditions. For example, the convolution kernel may identify the impact of temperature changes on read / write rate errors.

[0180] S2222, performing nonlinear mapping on the local features after the convolution operation through an activation function to obtain a feature map after nonlinear mapping;

[0181] It should be noted that the role of the activation function is to introduce nonlinear transformations so that the network can learn more complex patterns rather than simple linear relationships.

[0182] Common activation functions include ReLU (Rectified Linear Unit), Sigmoid, Tanh, etc.

[0183] ReLU is the most commonly used activation function, which converts all negative values to zero and keeps the positive values unchanged. The formula is:

[0184] ReLU( x )=max(0, x );

[0185] in, x Represents the value input to the activation function, typically the output of a node (neuron) in a neural network.

[0186] The ReLU activation function performs nonlinear mapping on the local features after the convolution operation, thereby increasing the network's expressive power. The feature map after nonlinear mapping will better reflect the complex relationships in the error data.

[0187] S2223. Use the channel attention mechanism to perform weighted processing on the feature map after nonlinear mapping to obtain a weighted feature map.

[0188] Specifically, the channel attention mechanism determines the importance of each channel by assigning different weights to different feature channels. It allows the network to focus on the most important feature channels in the prediction task.

[0189] After convolution and activation functions, multiple feature maps are obtained. Each feature map represents certain specific local features. Through the channel attention mechanism, a weight is assigned to each channel (i.e., each feature map). These weights represent the importance of that feature in the compensation task.

[0190] After calculating the attention weight of each channel, it is multiplied by each channel in the original feature map to obtain the weighted feature map.

[0191] S223. Based on the weighted feature map, a data-driven read / write rate compensation model is constructed through a fully connected layer.

[0192] Specifically, a weighted feature map is a tensor with multiple channels that represents the weighted results of different features. Before entering the fully connected layer, it is usually necessary to flatten the weighted feature map into a one-dimensional vector because the fully connected layer usually accepts a one-dimensional input vector.

[0193] For example, if the shape of the weighted feature map is ( H , W , C ), which after flattening becomes a H × W × C A one-dimensional vector of .

[0194] A fully connected layer transforms the input vector into an output using a set of weights and biases. Each neuron in a fully connected layer is connected to all neurons in the previous layer, so each output is a weighted sum of all inputs.

[0195] Assume the input vector size is H × W × C , then the size of the weight matrix of the fully connected layer is ( H × W × C , G ),in G is the output dimension of the fully connected layer (i.e., the target dimension predicted by the model, usually a single read / write rate value).

[0196] Assuming the input vector is C, the output of the fully connected layer is V, the weight is O, and the bias is b, the calculation formula of the fully connected layer is: ;

[0197] Among them, C is the flattened weighted feature map, O is the weight matrix, b is the bias vector, and V is the output of the fully connected layer, that is, the compensated read and write rate.

[0198] S3. Use the read / write rate compensation model to perform error correction on the preliminary prediction result of the solid-state drive read / write rate to obtain the final solid-state drive read / write rate.

[0199] Specifically, using the read / write rate compensation model to perform error correction on the preliminary prediction results of the solid-state drive read / write rate to obtain the final solid-state drive read / write rate includes the following steps:

[0200] The preliminary prediction results are used as input to the trained compensation model.

[0201] The model outputs the corrected read / write rate prediction value. After error correction, a more accurate prediction result of the SSD read / write rate can be obtained.

[0202] After obtaining the corrected read and write rates, these values can be used for multiple purposes, such as:

[0203] Evaluating SSD (Solid State Drive) Performance:

[0204] The error-corrected read and write rates can more accurately evaluate the actual performance of the solid-state drive, avoiding deviations caused by environmental changes, hard drive health status, and other factors.

[0205] Optimize system configuration:

[0206] Based on the revised read and write rate prediction results, the system configuration can be optimized, such as selecting the appropriate SSD model in the data center and adjusting the storage tier.

[0207] Choose the right SSD model:

[0208] The revised performance prediction can help users make more scientific choices among different SSD models to meet the needs of specific application scenarios.

[0209] like Figure 2 As shown, according to one embodiment of the present invention, a solid-state hard disk read and write rate analysis system is provided, including:

[0210] Preliminary prediction module 1 is used to collect the status information and read / write speed of the solid-state drive, analyze the influencing indicators of the solid-state drive read / write speed using the extension hierarchical analysis method, and make a preliminary prediction of the solid-state drive read / write speed through the fuzzy closeness calculation method;

[0211] Compensation model construction module 2 is used to perform joint simulation of solid-state drives based on digital twin technology and build a read and write rate compensation model for solid-state drives in combination with recurrent neural networks;

[0212] The read / write rate correction module 3 is used to perform error correction on the preliminary prediction result of the solid-state hard disk read / write rate using the read / write rate compensation model to obtain the final solid-state hard disk read / write rate.

[0213] In summary, with the help of the above technical solutions of the present invention, the present invention combines a variety of advanced technologies and methods and has strong adaptability and scalability. With the continuous development of solid-state drive technology and changes in the hardware environment, the accuracy and effectiveness of the method can be maintained by updating the digital twin model, adjusting the parameters of the recurrent neural network, etc., which can not only improve the prediction accuracy of the solid-state drive read and write rate, but also provide strong support for the performance evaluation, system optimization and hardware selection of the solid-state drive. The present invention uses the extension hierarchical analysis method to calculate the initial weights of each influencing indicator and corrects the initial weights through the ordinal relationship analysis method to obtain more reasonable final indicator weights. The data fitting technology is used to determine the fitting relationship between each influencing indicator and the solid-state drive read and write rate, which can accurately reflect the intrinsic relationship between the indicator and the read and write rate. The fuzzy closeness calculation method is used to make a preliminary prediction of the solid-state drive read and write rate, taking into account the fuzziness and similarity between samples. By calculating the closeness between the training sample and the prediction sample, and selecting a preset number of training samples for weighted average prediction, the prediction result is closer to the actual situation. The present invention uses digital twin technology to construct a virtual model of the solid-state drive, which can realize real-time simulation and emulation of the solid-state drive state. By integrating the attention mechanism, the convolutional neural network can pay more attention to the features that have a greater impact on the read and write rates, thereby improving the sensitivity and accuracy of the model. The introduction of the attention mechanism enables the model to dynamically adjust the degree of attention to different features, thereby enhancing the adaptability and robustness of the model. By calculating the difference between the simulated read and write rate data and the theoretical read and write rate data, the read and write rate error is obtained as the input of the compensation model. The data-driven compensation model can make full use of existing data and improve the accuracy and reliability of the prediction.

[0214] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0215] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing the read and write speed of a solid-state hard disk, characterized in that: include: S1. Collect the status information and read / write speed of the solid-state drive, use the extension analytic hierarchy process to analyze the influencing indicators of the solid-state drive read / write speed, and make a preliminary prediction of the solid-state drive read / write speed through the fuzzy proximity calculation method; S2. Based on digital twin technology, we conduct joint simulation of solid-state drives and build a read and write rate compensation model for solid-state drives using a recurrent neural network. include: S21. Use digital twin technology to build a virtual model of the solid-state drive, and use the virtual model of the solid-state drive to perform joint simulations under different conditions to obtain simulated read and write rate data; S22. Combining theoretical and simulated read / write rate data of solid-state drives, and using a convolutional neural network with an attention mechanism, we build a data-driven read / write rate compensation model, including: S221. Determine theoretical read / write rate data of the solid-state drive under the same conditions based on the simulated read / write rate data of the solid-state drive, and calculate the difference between the simulated read / write rate data and the theoretical read / write rate data to obtain a read / write rate error. S222. Using a convolutional neural network, extract local features using the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data, and perform weighted processing on the extracted local features using a channel attention mechanism to generate a weighted feature map. S223. Based on the weighted feature map, a data-driven read / write rate compensation model is constructed through a fully connected layer. S3. Using the read / write rate compensation model, perform error correction on the preliminary prediction result of the solid-state drive read / write rate to obtain the final solid-state drive read / write rate; Said S1 comprises: Based on the read and write speeds of solid-state drives, data fitting technology is used to determine the fitting relationship between various influencing indicators and the read and write speeds of solid-state drives. Combined with the final indicator weights, a fuzzy proximity calculation method is used to make a preliminary prediction of the read and write speeds of solid-state drives. Among them, the method of combining the final indicator weight and using the fuzzy proximity calculation method to make a preliminary prediction of the read and write rate of the solid-state hard drive includes: establishing training samples based on the fitting relationship between each indicator and the read and write rate of the solid-state hard drive, calculating the proximity between the training samples and the predicted samples in combination with the final indicator weight, and making a preliminary prediction of the read and write rate through the weighted average method.

2. A solid state hard disk read and write rate analysis method according to claim 1, characterized in that: The method of using data fitting technology to determine the fitting relationship between each influencing indicator and the read / write speed of the solid-state drive based on the read / write speed of the solid-state drive, and combining the final indicator weights with the fuzzy proximity calculation method to make a preliminary prediction of the read / write speed of the solid-state drive includes: The state information of the solid-state drive is used as different influencing indicators that affect the read and write speed of the solid-state drive, and the extension interval number judgment matrix is constructed by comparing the indicators pairwise. Based on the extension interval number judgment matrix, the extension hierarchical analysis method is used to calculate the initial weights of each influencing indicator, and the initial weights are corrected through the order relationship analysis method to obtain the final indicator weights.

3. The method for analyzing the read and write speed of a solid-state hard disk according to claim 2, wherein: The initial weights of the influencing indicators are calculated based on the extension interval number judgment matrix using the extension hierarchical analysis method, and the initial weights are corrected by the order relationship analysis method to obtain the final indicator weights, including: S121, performing a consistency index test on the extension interval number judgment matrix, and calculating the normalized eigenvector corresponding to the maximum eigenroot of the extension interval number judgment matrix that passes the consistency test to obtain a normalized weight vector; S122. Form a weight interval vector that satisfies the consistency condition based on the normalized weight vector, and calculate the single ranking weight of each influencing indicator corresponding to its upper-level indicator based on the weight interval vector; S123, calculating the group decision average indicator weight vector based on the single ranking weight to obtain the initial weight; S124. Use the ordinal relationship analysis method to determine the importance ranking of the subjective weights and calculate the relative importance ratio. According to the importance ranking of the subjective weights and the relative importance ratio, modify the initial weights to obtain the final indicator weights.

4. The method for analyzing the read and write speed of a solid-state hard disk according to claim 2, wherein: The method of using data fitting technology to determine the fitting relationship between each influencing indicator and the read / write rate of the solid-state hard disk based on the read / write rate of the solid-state hard disk includes: S131, preprocessing the read and write rates of the solid-state drive, wherein the preprocessing includes: removing noise, filling missing values, and normalizing; S132. Use nonlinear fitting to fit each influencing indicator and the read / write rate to obtain a fitting relationship between each indicator and the read / write rate of the solid-state drive.

5. A method for analyzing the read and write speed of a solid-state hard disk according to claim 4, characterized in that: The method of establishing training samples based on the fitting relationship between each indicator and the read / write rate of the solid-state drive, calculating the closeness between the training samples and the prediction samples in combination with the final indicator weight, and making a preliminary prediction of the read / write rate by the weighted average method includes: S1331, using the pre-processed read / write rate as the sample output and the read / write rate influencing index as the sample input to form a training sample matrix; S1332. Obtain prediction samples of the read / write rates of the solid-state drive to be predicted, and calculate the closeness between the training samples and the prediction samples using a closeness calculation formula based on the final indicator weights; S1333. Sort the obtained closeness, select a preset number of training samples according to the sorting result, and predict the read and write rates of the solid-state drive to be tested by a weighted average method to obtain a preliminary prediction result.

6. The method for analyzing the read and write speed of a solid-state hard disk according to claim 5, wherein: The closeness calculation formula is: ； ； ； Where, H ( T , T ′) represents the closeness, T i Indicates the i training samples, T ′ represents the sample to be predicted, Expressed as the union operation of fuzzy sets, It is represented as the intersection operation of fuzzy sets, Y represents the set of read and write rates corresponding to all influencing indicators, y ki Indicates the i Of the training samples k The reading and writing rates corresponding to the impact indicators are: y′ k Indicates the first k The reading and writing rates corresponding to the impact indicators are: w k Indicates the k The final indicator weight of each influencing indicator.

7. The method for analyzing the read and write speed of a solid-state hard disk according to claim 6, wherein: The method of extracting local features by using the read / write rate error between the simulated read / write rate data and the theoretical read / write rate data through a convolutional neural network, and weighting the extracted local features by using a channel attention mechanism to generate a weighted feature map includes: S2221, inputting the read / write rate error into a convolutional neural network, and extracting local features from the input read / write rate error data through a convolution operation; S2222, performing nonlinear mapping on the local features after the convolution operation through an activation function to obtain a feature map after nonlinear mapping; S2223. Use the channel attention mechanism to perform weighted processing on the feature map after nonlinear mapping to obtain a weighted feature map.

8. A solid-state hard disk read / write rate analysis system, used to implement the solid-state hard disk read / write rate analysis method according to any one of claims 1 to 7, characterized in that: include: The preliminary prediction module is used to collect the status information and read and write speed of the solid-state drive, analyze the influencing indicators of the solid-state drive read and write speed using the extension hierarchical analysis method, and make a preliminary prediction of the solid-state drive read and write speed through the fuzzy proximity calculation method; A compensation model building module is used to co-simulate solid-state drives based on digital twin technology and build a read and write rate compensation model for solid-state drives using a recurrent neural network. The read / write rate correction module is used to use the read / write rate compensation model to perform error correction on the preliminary prediction results of the solid-state hard disk read / write rate to obtain the final solid-state hard disk read / write rate.

Images