Cross-level data prefetching method and device, equipment and medium

By monitoring the changes in data access mode in real time and using the target data access frequency prediction model to migrate data to cache, the problem of low cache hit rate in the complex data access mode of traditional cache management strategies is solved, and system performance and response speed is improved.

CN119003387APending Publication Date: 2024-11-22INSPUR SUZHOU INTELLIGENT TECH CO LTD
View PDF 0 Cites 7 Cited by

Patent Information

Application Number
CN202411214840.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When facing complex and dynamic data access modes, traditional cache management strategies are difficult to accurately predict and quickly adapt to changes in access modes, resulting in a reduced cache hit rate, increasing data access latency, affecting system performance and response speed.

Method used

By monitoring the changes in data access mode in real time, using the target data access frequency prediction model to predict the access frequency, the target data is pre-migrated from memory to cache based on the prediction results, so as to quickly respond to data access requests. The model combines system status, context information and time information to update the initial model weight parameters and predicts through long-term and short-term memory neural networks.

Benefits of technology

Significantly improves cache hit rate, reduces data access latency, improves system performance and response speed, and improves user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119003387A_ABST
    Figure CN119003387A_ABST
Patent Text Reader

Abstract

The invention discloses a cross-level data prefetching method and device, equipment and a medium, and relates to the technical field of data access, and the method comprises the steps: monitoring whether a data access mode related to a business activity is changed or not in real time after the business activity is updated; if the data access mode changes, predicting an access frequency in the data access mode by using a target data access frequency prediction model of a preset data deployment management system to obtain a corresponding access frequency prediction result; and according to the access frequency prediction result, predicting target data which needs to be accessed in the next access time and is located in the memory, and migrating the predicted target data from the memory to a preset cache in advance, so that the preset cache is utilized to respond to a data access request initiated for the target data in the next access time. According to the method, the target data of the next access time is accurately predicted through the target data access frequency prediction model, and the target data is migrated to the cache, so that cross-level data prefetching is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data access technology, and in particular to a cross-level data pre-fetching method, device, equipment and medium. Background Art

[0002] In modern computing systems, cache is a key component to improve data access speed and system performance. However, traditional cache management strategies, such as LRU (Least Recently Used) and LFU (Least Frequently Used), often expose certain limitations when dealing with complex, changing and dynamic data access patterns. Traditional cache management strategies are difficult to accurately predict and quickly adapt to changes in access patterns when faced with complex and dynamic data access patterns. Since traditional strategies cannot effectively deal with complex access situations, the data stored in the cache does not match the actual access requirements, thereby reducing the cache hit rate. The reduction in cache hit rate means that more data needs to be obtained from slower storage levels, increasing access latency, and ultimately having a negative impact on the performance and response speed of the system, affecting the performance and response speed of the system, and further leading to problems such as slow system operation and poor user experience.

[0003] It can be seen that when faced with complex and dynamic data access patterns, how to accurately adapt to changes in access patterns in the system and timely process and match corresponding cache data is a problem that technical personnel in this field need to solve. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a cross-level data pre-fetching method, device, equipment and medium, which can accurately adapt to changes in access patterns in the system and timely process and match corresponding cache data.

[0005] In a first aspect, the present invention discloses a cross-level data pre-fetching method, which is applied to a preset data allocation management system, comprising:

[0006] After a business activity is updated, monitor in real time whether the data access pattern related to the business activity has changed;

[0007] If the data access mode changes, the target data access frequency prediction model of the preset data allocation and management system is used to predict the access frequency under the data access mode to obtain a corresponding access frequency prediction result; wherein, the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target corresponding relationship, and the target corresponding relationship is the corresponding relationship between each system state of the preset data allocation and management system under different data access modes, context information related to data access under different data access modes, time information corresponding to the execution of different data access modes, and the access frequency of the accessed data under different data access modes;

[0008] Based on the access frequency prediction result, the target data located in the memory that needs to be accessed within the next access time is predicted, and the predicted target data is pre-migrated from the memory to a preset cache, so that the preset cache can be used to respond to the data access request initiated for the target data within the next access time.

[0009] Optionally, before predicting the access frequency under the data access mode using the target data access frequency prediction model of the preset data allocation management system, the method further includes:

[0010] Collect the historical access frequencies of the corresponding historically accessed data under different data access modes of the preset data allocation management system, the historical system status of the preset data allocation management system under the corresponding data access mode, the historical context information related to the data access under the corresponding data access mode, and the historical time information corresponding to the execution of the corresponding data access mode, so as to obtain a data training set;

[0011] The initial data access frequency prediction model is trained using the data training set to update the weight parameters of the initial data access frequency prediction model to obtain a target data access frequency prediction model; wherein the initial data access frequency prediction model is an initial model constructed based on a system state calculation factor, a weight parameter of the system state calculation factor, an access frequency calculation factor of accessed data, a weight parameter of the access frequency calculation factor, a context feature factor, a weight parameter of the context feature factor, a time feature factor, and a weight parameter of the time feature factor.

[0012] Optionally, the using the data training set to perform model training on the initial data access frequency prediction model to update each weight parameter of the initial data access frequency prediction model to obtain a target data access frequency prediction model includes:

[0013] The data training set is input into an initial data access frequency prediction model including an input layer, a long short-term memory neural network layer, a fully connected layer and an output layer connected in sequence, so that the historical feature vector obtained after the data training set is processed by the input layer is processed by the long short-term memory neural network layer of the initial data access frequency prediction model according to the cell state update rule and the hidden state update rule, so as to adjust and obtain the corresponding hidden state according to the historical time information, and then the hidden state is mapped in a high-dimensional manner through the fully connected layer to obtain the mapped data, and the mapped data is adjusted in importance and globally adjusted through the output layer to output the predicted access frequency of the historical accessed data;

[0014] The weight parameters in the initial data access frequency prediction model are adjusted according to the predicted access frequency and the corresponding historical access frequency of the historically accessed data to obtain a target data access frequency prediction model.

[0015] Optionally, before the long short-term memory neural network layer of the initial data access frequency prediction model processes the historical feature vector obtained after the data training set is processed by the input layer according to the cell state update rule and the hidden state update rule, it also includes:

[0016] The data training set is input into the input layer of the initial data access frequency prediction model so that the input layer constructs a historical feature vector under the corresponding data access mode based on the historical system status under different data access modes in the data training set, the historical context information related to data access under the corresponding data access mode, and the historical access frequency of the corresponding historical accessed data when the corresponding data access mode is executed.

[0017] Optionally, the pre-migrating the predicted target data from the memory to a preset cache includes:

[0018] Acquire a first priority storage strategy of the target data at the current moment; the first priority storage strategy is a storage strategy of a target storage space in the memory storing the target data;

[0019] Execute the step of updating the first priority storage strategy of the target data to a second priority storage strategy, and obtain the second priority storage strategy of the target data, so as to pre-migrate the target data from the target storage space of the memory to the first-level cache of the central processing unit;

[0020] Among them, the second priority storage strategy is a storage strategy for storing the target data in the first-level cache of the central processing unit; the data access speed in the second priority storage strategy is higher than the data access speed in the first priority storage strategy.

[0021] Optionally, before obtaining the first priority storage strategy of the target data at the current moment, the method further includes:

[0022] If the number of data accesses to each accessed data under the historical access time is less than a preset access number threshold, a first priority storage strategy is constructed to characterize the storage space of each accessed data stored in the memory under the historical access time according to the data heat, data access frequency, data freshness and system load of each accessed data and the corresponding parameter coefficients; wherein the data freshness is the inverse of the historical access time of the accessed data;

[0023] Migrating each of the accessed data to a corresponding storage space in the memory according to the first priority storage strategy, generating a data cache strategy for each storage space in the memory and setting a corresponding data cache size;

[0024] If the number of data accesses for each accessed data at the historical access time is greater than or equal to a preset access number threshold, a second priority storage strategy is constructed according to the data heat, data access frequency, data freshness and system load of each accessed data and the corresponding parameter coefficients thereof to store each accessed data at the historical access time in the cache of the central processing unit;

[0025] When receiving an access request for the accessed data, the accessed data is extracted from the corresponding storage location through the preset data allocation management system; wherein the data cache size and data cache strategy of the cache and the storage space are adjusted according to each of the access frequency prediction results.

[0026] Optionally, the cross-level data pre-fetching method further includes:

[0027] Obtaining the storage cost of the memory, the storage cost in the central processor, the system load cost of the preset data allocation management system occupied by the access frequency prediction results of each accessed data predicted by the target data access frequency prediction model, and the time cost generated by obtaining the accessed data from the storage location, so as to obtain the cache strategy adjustment factor;

[0028] The cache policy adjustment factor is used to adjust the storage space of the memory and the data cache size and data cache policy of the cache of the central processing unit.

[0029] In a second aspect, the present invention discloses a cross-level data pre-fetching device, which is applied to a preset data allocation management system, comprising:

[0030] A monitoring module, used to monitor in real time whether a data access mode related to a business activity changes after the business activity is updated;

[0031] A prediction module, which is used to predict the access frequency under the data access mode using the target data access frequency prediction model of the preset data allocation and management system if the data access mode changes, and obtain a corresponding access frequency prediction result; wherein the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target corresponding relationship, and the target corresponding relationship is the corresponding relationship between each system state of the preset data allocation and management system under different data access modes, the context information related to data access under different data access modes, the time information corresponding to the execution of different data access modes and the access frequency of the accessed data under different data access modes;

[0032] A data prefetch module is used to predict the target data located in the memory that needs to be accessed within the next access time based on the access frequency prediction result, and to pre-migrate the predicted target data from the memory to a preset cache so that the preset cache can be used to respond to the data access request initiated for the target data within the next access time.

[0033] In a third aspect, the present invention discloses an electronic device, comprising:

[0034] Memory for storing computer programs;

[0035] A processor is used to execute the computer program to implement the steps of the cross-level data prefetching method disclosed above.

[0036] In a fourth aspect, the present invention discloses a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the cross-level data prefetching method disclosed above are implemented.

[0037] To solve the above technical problems, an embodiment of the present invention provides a cross-level data pre-fetching method, which is applied to a preset data allocation and management system, including: after a business activity is updated, real-time monitoring of whether the data access mode related to the business activity has changed; if the data access mode has changed, using the target data access frequency prediction model of the preset data allocation and management system to predict the access frequency under the data access mode, and obtaining a corresponding access frequency prediction result; wherein the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target correspondence relationship, and the target correspondence relationship is the correspondence between each system state of the preset data allocation and management system under different data access modes, context information related to data access under different data access modes, time information corresponding to the execution of different data access modes, and the access frequency of the accessed data under different data access modes; according to the access frequency prediction result, the target data located in the memory that needs to be accessed within the next access time is predicted, and the predicted target data is pre-migrated from the memory to the preset cache, so as to use the preset cache to respond to the data access request initiated for the target data within the next access time.

[0038] It can be seen from the above technical solutions that by real-time monitoring of changes in data access patterns, the dynamic adjustment of data access behavior can be captured in a timely manner, no longer limited to the fixed patterns of traditional strategies, and thus better adapted to complex and changeable data access situations. Utilizing the target data access frequency prediction model, the access frequency is predicted by comprehensively considering multiple factors such as system status, context information, and time information. Compared with traditional strategies, it can more accurately predict data requirements, thereby improving the matching degree between data in the cache and actual access requirements, and significantly improving the cache hit rate. Determine the target access data based on accurate prediction results and migrate it to the internal memory. When the system initiates an access request, it can extract data from the fast internal memory, greatly reducing data access delays, significantly improving system performance and response speed, and improving user experience. The system performance is optimized and the response speed is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention, 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 ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a cross-level data pre-fetching method provided by an embodiment of the present invention;

[0041] Figure 2A flow chart of a storage strategy construction and storage strategy update method provided by an embodiment of the present invention;

[0042] Figure 3 A flow chart of a cross-level data pre-fetching method after a data access mode change provided by an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of the structure of a cross-layer data pre-fetching device provided by an embodiment of the present invention;

[0044] Figure 5 A structural diagram of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The terms "including" and "having" in the specification of the present invention and the above-mentioned drawings, as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

[0047] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0048] In modern computing systems, cache is a key component to improve data access speed and system performance. However, traditional cache management strategies (such as LRU and LFU) often expose certain limitations when dealing with complex, changeable and dynamic data access patterns. Traditional cache management strategies are difficult to accurately predict and quickly adapt to changes in access patterns when faced with complex and dynamic data access patterns. Since traditional strategies cannot effectively deal with complex access situations, the data stored in the cache does not match the actual access requirements, thereby reducing the cache hit rate. The reduction in cache hit rate means that more data needs to be obtained from slower storage levels, increasing access latency, and ultimately having a negative impact on the performance and response speed of the system, affecting the performance and response speed of the system, and further leading to problems such as slow system operation and poor user experience.

[0049] To this end, the present invention provides a cross-level data prefetching solution that can accurately adapt to changes in access patterns in the system when faced with complex and dynamic data access patterns, and process and match corresponding cache data in a timely manner.

[0050] Reference Figure 1 As shown, the present invention provides a cross-level data pre-fetching method, which is applied to a preset data allocation management system, comprising:

[0051] Step S11: After the business activity is updated, real-time monitoring is performed to determine whether the data access mode associated with the business activity has changed.

[0052] In this embodiment, when different business activities are initiated or updated, the data access mode related to the business activities is monitored in real time to see whether it changes. For example, in a webpage website on the browser side, the organizer displays that a business activity has been initiated on the front end of the webpage. From the user's perspective, it is easier to click on the activity page when seeing a new activity. Correspondingly, the background will call the data related to the business activity from the corresponding storage location. Because the entire process of calling the data involves the storage data located in different storage locations and different storage spaces, the access request volume of the data related to the business activity will increase greatly, and the data access mode will change accordingly.

[0053] It should be noted that the business operations generated by users when participating in business activities involve calling different data, and different data have corresponding storage locations according to the previous storage strategy. When the data storage location called is in an external storage device such as a hard disk, the reading speed is slow. In addition, because the organizer sets up business activities, there may be a situation where multiple users call the same / same batch of data stored in the hard disk very frequently in a short period of time. In this case, each call to the same data will waste a lot of time. At the same time, the preset data allocation management system monitors in real time whether the data access pattern related to the updated business activities has changed. For example: in an e-commerce website, monitoring the data access pattern can find that during a specific promotional event, the detailed information and user reviews of certain products are frequently accessed; on a video streaming platform, it may be found that a specific type of video has a higher access frequency in a specific time period (such as weekend nights). If the data access pattern changes, it means that the updated business activities affect the number of times the background data is accessed and read.

[0054] Furthermore, in order to more accurately monitor which data access mode or modes have changed, multiple access mode change types are predefined:

[0055] 1. Sudden increase in access frequency: Set a threshold. When the access frequency of a specific data block or data type exceeds this threshold within a certain period of time, it is determined that the access frequency has suddenly increased.

[0056] 2. Concentrated access to specific types of data: By analyzing the attributes or labels of the data, specific types of data are identified. When these specific types of data are accessed in a concentrated manner within a period of time, it is determined to be concentrated access to specific types of data.

[0057] 3. Re-access of data that has not been accessed for a long time: Track the access history of data. When data that has not been accessed for a long time is suddenly accessed, it is determined as a re-access of the data that has not been accessed for a long time.

[0058] Establish multi-dimensional monitoring indicators:

[0059] In addition to the traditional number of accesses, other indicators such as access time interval, access duration, and concurrent access volume can also be considered. For example, when multiple users access the same data block at the same time, the concurrent access volume increases, which may be an important change signal. Different monitoring indicator weights can be set for different types of data. For example, for data with high real-time requirements, the access time interval may be more important; for large-capacity data, the concurrent access volume may be more worthy of attention.

[0060] In this embodiment, a time series analysis method is used to predict the future trend of the data access mode. By analyzing the change trend of historical data, possible changes in access modes can be discovered in advance, providing earlier warnings for the intelligent pre-fetching algorithm module and the cross-level data migration module. Or, a machine learning algorithm is used to perform anomaly detection to identify abnormal access behaviors that are different from normal access modes. For example, a clustering algorithm is used to divide data access modes into different categories. When an access mode that does not belong to any known category appears, it is determined to be an abnormal access. It can be seen that by classifying and monitoring the types of access mode changes, the intelligent pre-fetching algorithm module and the cross-level data migration module can obtain more specific information, thereby making predictions and data migration more targeted. For example, when a sudden increase in access frequency is detected, the intelligent pre-fetching algorithm can give priority to predicting the future access needs of these data and pre-fetch them into a high-level cache; when concentrated access to a specific type of data is detected, the cross-level data migration module can consider migrating the relevant data from a slower level of memory to a faster level of cache to meet the needs of concentrated access.

[0061] Step S12: If the data access mode changes, the target data access frequency prediction model of the preset data allocation and management system is used to predict the access frequency under the data access mode to obtain a corresponding access frequency prediction result; wherein, the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target correspondence relationship, and the target correspondence relationship is the correspondence between the system states of the preset data allocation and management system under different data access modes, the contextual information related to data access under different data access modes, the time information corresponding to the execution of different data access modes, and the access frequency of the accessed data under different data access modes.

[0062] In this embodiment, before using the target data access frequency prediction model of the preset data allocation management system to predict the access frequency under the data access mode, it also includes: collecting the historical access frequency of the corresponding historical accessed data under different data access modes of the preset data allocation management system and the historical system state of the preset data allocation management system under the corresponding data access mode, the historical context information related to data access under the corresponding data access mode, and the corresponding historical time information when executing the corresponding data access mode to obtain a data training set; using the data training set to perform model training on the initial data access frequency prediction model to update the weight parameters of the initial data access frequency prediction model to obtain a target data access frequency prediction model; wherein the initial data access frequency prediction model is an initial model constructed based on a system state calculation factor, a weight parameter of the system state calculation factor, an access frequency calculation factor of the accessed data, a weight parameter of the access frequency calculation factor, a context feature factor, a weight parameter of the context feature factor, a time feature factor, and a weight parameter of the time feature factor. It can be understood that in order to accurately predict the target data that may be accessed within the next access time after the data access mode changes, and then move the target data to the cache in advance, so that when receiving a data access request for the target data, the corresponding target data can be directly retrieved from the cache for response, by pre-training a target data access frequency prediction model that can accurately predict the access frequency of the access data after the data access mode changes, to predict the access frequency under the corresponding data access mode after the data access mode changes. Therefore, the historical access frequency of the corresponding historical access data under different data access modes of the preset data allocation management system and the historical system state of the preset data allocation management system under the corresponding data access mode, the historical context information related to data access under the corresponding data access mode, and the corresponding historical time information when the corresponding data access mode is executed are collected to obtain a data training set. The data training set is then used to train the initial data access frequency prediction model, wherein the initial data access frequency prediction model is represented as follows:

[0063] ;

[0064] in, , , , They are the weight parameters corresponding to the system status calculation factor, the access frequency calculation factor of the access data, the context feature factor, and the time feature factor. Represents an impact score calculated based on historical system status. That is, the system state calculation factor, if If it is being requested frequently, the impact score is high; Indicates the historical access frequency calculated based on historical access data. That is, the access frequency calculation factor, where the historical access frequency is determined by the historical access count and trend analysis of the historical access data; Indicates the degree of influence of historical context information related to any data access. That is, the contextual feature factor, Represents time characteristic information related to data access, Represents the time characteristic factor. represents the target historical access frequency under the data access mode corresponding to the above input feature variables. It should be noted that , , , The weight parameters represented are not the weight parameters of the final target model, and the initial model needs to be continuously trained and updated by the data training set to be determined.

[0065] The specific training process is: inputting the data training set into an initial data access frequency prediction model including an input layer, a long short-term memory neural network layer, a fully connected layer and an output layer connected in sequence, so that the historical feature vector obtained after the data training set is processed by the input layer is processed according to the cell state update rule and the hidden state update rule by the long short-term memory neural network layer of the initial data access frequency prediction model, so as to adjust and obtain the corresponding hidden state according to the historical time information, and then perform high-dimensional mapping on the hidden state through the fully connected layer to obtain the mapped data, and perform importance adjustment and global adjustment on the mapped data through the output layer to output the predicted access frequency of the historically accessed data; according to the predicted access frequency and the corresponding historical access frequency of the historically accessed data, each weight parameter in the initial data access frequency prediction model is adjusted to obtain the target data access frequency prediction model. Among them, before the long short-term memory neural network layer of the initial data access frequency prediction model processes the historical feature vector obtained after the data training set is processed by the input layer according to the cell state update rule and the hidden state update rule, it also includes: inputting the data training set into the input layer of the initial data access frequency prediction model, so that the input layer constructs the historical feature vector under the corresponding data access mode based on the historical system state under different data access modes in the data training set, the historical context information related to data access under the corresponding data access mode, and the historical access frequency of the corresponding historical accessed data when the corresponding data access mode is executed.

[0066] It can be understood that the historical feature vector constructed by the input layer based on the input information is: X input = [ x historical , x current , x context ] ,in, , , They respectively represent historical access data (including data access records and trends over the past period of time), historical system status (the status of ongoing data access activities in the current system, such as which data is being requested), and historical context information (covering factors such as time, location, and environment that may affect data access frequency). Indicates that based on the above , , The constructed historical feature vector then enters the LSTM (Long Short-Term Memory) layer of the long short-term memory neural network layer, and the LSTM layer updates the rules: 1. The cell state update rule is that the cell state is obtained by multiplying the output of the forget gate by the cell state at the previous moment element by element, and then adding the output of the input gate by the candidate cell state element by element. The cell state update rule is specifically expressed as follows: ;

[0067] in, Indicates the cell state corresponding to the current input, is the output of the forget gate, represents an element-wise multiplication operation, is the output of the input gate, is a candidate cell state.

[0068] Thus, during the initial model training, the training process based on the cell state update rule includes: deciding which information to discard from the cell state based on the current input and the hidden state at the previous moment. It adjusts the cell state by learning long-term dependencies and unimportant features in the data. For example, if historical data indicates that the access pattern of some data has changed, the forget gate may reduce the memory of relevant information in the past. The input gate and candidate cell states determine which new information needs to be added to the cell state. The input gate filters out important new information based on the current input and the hidden state at the previous moment. The candidate cell state is a preliminary processing of the new information. By continuously adjusting the input gate and candidate cell states, the model can learn how to update the cell state to adapt to new data input.

[0069] Finally, the cell state is updated through the above cell state update rules, retaining important historical information while incorporating new information.

[0070] 2. Hidden state update rule, that is, the hidden state of the LSTM layer at time step t is obtained by multiplying the output gate of the LSTM layer and the result of the hyperbolic tangent activation function acting on the cell state element by element, expressed as: ;

[0071] in, is the output of the output gate, and tanh is the hyperbolic tangent activation function.

[0072] In this way, the output gate decides which information should be output from the cell state as the hidden state. It selects information that is useful for the current prediction based on the current input, the hidden state at the previous moment, and the cell state. The hidden state is updated using the above hidden state update rule, where the tanh activation function performs a nonlinear transformation on the cell state so that the hidden state can better represent the characteristics of the data. The hidden state contains a summary of past data and a response to the current input, which will be passed to the next layer for further processing.

[0073] Furthermore, the representation of the fully connected layer Dense is as follows:

[0074] ;

[0075] in, is the weight matrix of the fully connected layer, which is used to weight different features of the hidden state and learn the importance of different features in prediction. Represents a bias vector, which is used to provide an offset for the linear transformation. Represents the hidden state at time step t. ReLU is the activation function, defined as x↦max(0,x). The ReLU activation function performs a nonlinear transformation on the result of the linear transformation, setting values ​​less than 0 to 0 and retaining values ​​greater than 0. This enables the model to learn more complex feature representations and enhance the model's expressiveness.

[0076] The output layer is represented as follows:

[0077] ;

[0078] in, is the weight matrix of the output layer, is the bias term of the output layer.

[0079] By continuously inputting the data training set into the initial data access frequency prediction model, the parameters in each layer structure of the initial data access frequency prediction model are adjusted using the data training set, especially the weight matrix and the bias term During the training process, adjustments are made continuously to make the predicted access frequency as close to the actual access frequency as possible. Finally, a loss function is used to measure the difference between the predicted access frequency and the actual access frequency. Common loss functions include mean squared error. Through optimization algorithms (such as stochastic gradient descent), various parameters of the model (including the gating parameters of the LSTM layer, the weight matrix and bias vector of the fully connected layer, the weight matrix and bias term of the output layer, etc.) are continuously adjusted to minimize the loss function. The optimization algorithm gradually updates the parameters based on the gradient information of the loss function, making the prediction results of the model more and more accurate.

[0080] In this embodiment, after training the target data access frequency prediction model, although the above steps detect the change of data access pattern, it does not mean that the specific data to be accessed is completely known. Detecting the change of access pattern mainly detects the change of trend or law of data access behavior, but cannot accurately determine which specific data will be accessed next. Therefore, the access frequency after the data access pattern changes is predicted by the target data access frequency prediction model to obtain the corresponding access frequency prediction result.

[0081] Step S13: predict the target data located in the memory that needs to be accessed within the next access time based on the access frequency prediction result, and pre-migrate the predicted target data from the memory to a preset cache, so as to use the preset cache to respond to the data access request initiated for the target data within the next access time.

[0082] In this embodiment, the target data that needs to be accessed within the next access time is predicted based on the access frequency prediction result output by the model. It should be noted that the currently predicted target data is the data located in the memory, that is, the data that needs to perform a data pre-fetching step in advance. As for the target located in the cache (CPU L1 cache) that needs to be accessed within the next access time, since its current storage location is already in the cache, it is not used as the target data.

[0083] In this embodiment, the first priority storage strategy of the target data at the current moment is obtained; the first priority storage strategy is the storage strategy of the target data in the target storage space of the memory; the step of updating the first priority storage strategy of the target data to the second priority storage strategy is performed to obtain the second priority storage strategy of the target data, so as to pre-migrate the target data from the target storage space of the memory to the first-level cache of the central processing unit; wherein the second priority storage strategy is the storage strategy of the target data stored in the first-level cache of the central processing unit; the data access speed in the second priority storage strategy is higher than the data access speed in the first priority storage strategy. It can be understood that, since the current storage location of the target data to be accessed at the next access time is predicted to be located in a certain storage space (target storage space) of the memory, the first priority storage strategy of the target data is updated at the current moment to obtain the second priority storage strategy of the target data in the first-level cache of the central processing unit, and the step of updating the priority storage strategy is performed to migrate the target data to the first-level cache, and when the next access time arrives, the access request from the target data is received and responded.

[0084] In this embodiment, before obtaining the first priority storage strategy for the target data at the current moment, it also includes: if the number of data accesses to each accessed data at the historical access time is less than a preset access number threshold, then constructing a first priority storage strategy for the storage space of each accessed data stored in the memory at the historical access time according to the data heat, data access frequency, data freshness and system load of each accessed data and their corresponding parameter coefficients; wherein the data freshness is the inverse of the historical access time of the accessed data; according to the first priority storage strategy, each accessed data is migrated to the corresponding storage space in the memory, and a storage space of each storage space in the memory is generated. Data cache strategy and setting of corresponding data cache size; if the number of data accesses for each accessed data under the historical access time is greater than or equal to the preset access number threshold, then a second priority storage strategy is constructed to characterize the storage of each accessed data in the cache of the central processor under the historical access time according to the data heat, data access frequency, data freshness and system load of each accessed data and the corresponding parameter coefficients; when receiving an access request for the accessed data, the accessed data is extracted from the corresponding storage location through the preset data allocation management system; wherein the data cache size and data cache strategy of the cache and the storage space are adjusted according to the prediction results of each access frequency. It can be understood that the preset access number threshold is the average value of the corresponding number of data access changes when different historical access modes change. Then, the corresponding priority storage strategy can be pre-constructed according to the data heat, access frequency and other data priority information related to the access data of each data. Specifically, Figure 2 As shown, a low-priority storage strategy and a high-priority storage strategy are constructed according to the space size of the first-level, second-level, and third-level caches in the system's central processor, the amount of storage data to be stored, and the data priority information. The data storage priority is expressed as follows:

[0085] ;

[0086] in, , , , They are the weight coefficient of data heat, the weight coefficient of access frequency, the weight coefficient of data freshness, and the weight coefficient of system load.

[0087] In this embodiment, the storage cost of the memory, the storage cost in the central processing unit, the system load cost of the preset data allocation and management system occupied by the access frequency prediction results of each accessed data predicted by the target data access frequency prediction model, and the time cost generated by obtaining the accessed data from the storage location are obtained to obtain the cache strategy adjustment factor; the cache strategy adjustment factor is used to adjust the storage space of the memory and the data cache size and data cache strategy of the cache of the central processing unit. It can be understood that in order to more accurately realize data prefetching, the constructed high-priority storage strategy and low-priority storage strategy are changed over time and the retrograde strategy is updated, that is, the storage strategy update of the target data is completed over time, based on the storage strategy change of the target data, check and update whether the storage strategy of other data is updated accordingly. Specifically, by inputting characteristic variables into the target data access frequency prediction model X input = [ x current , x historical , x context ] ; Use function g to predict cache blocks based on input features The miss rate of the prediction is expressed as follows:

[0088] ;

[0089] After predicting the miss rate, we further construct the miss penalty, which is expressed as follows:

[0090] ;

[0091] By analyzing the input features, the model can predict the probability of a miss when a cache block is not in the current cache. This prediction can help determine which cache blocks are more likely to be accessed, thereby deciding whether to migrate them to a higher level cache. The penalty cost for a miss is determined based on the cache block location. The miss penalty takes into account the cost of fetching data from a slower level of memory when the data is not in the cache. The penalty cost includes storage cost and energy cost.

[0092] Storage costs: ;

[0093] in, Represents a cache block Storage Cost, Represents a cache block The size of the block (BlockSize), Indicates the cost per unit storage (Cost Per Unit Storage).

[0094] Energy cost: ;

[0095] in, Represents a cache block Energy Cost, It represents the energy consumption per access (Energy Per Access), Indicates a cache block Access Frequency

[0096] Total cost calculation: ;

[0097] in, Represents the total cost (Total Cost), Represents a cache block MissPenalty, Indicates a cache block The predicted miss rate (Predicted Miss Rate) is taken into account. By considering the miss penalty, storage cost, and energy cost, by combining these cost factors, it can be determined whether it is cost-effective to migrate data to a higher-level cache.

[0098] In this way, when a change in data access patterns is detected, data is migrated from the slower tier of storage to the faster tier of cache. The purpose is to increase data access speed and optimize system performance. By timely migrating data that may be frequently accessed to a faster storage tier, data access latency can be reduced and system response speed can be improved. By comprehensively considering storage costs and energy costs, it is ensured that data storage strategies and data migration will not consume excessive system resources while improving performance. By rationally selecting the data blocks to be migrated, optimal use of resources can be achieved.

[0099] Reference Figure 3 As shown, an overall method for data prefetching is disclosed, including: after the system is started, it enters the data access monitoring state. The data access mode is monitored in real time by the data access monitoring module of the system to determine whether there is a change. If there is no change, the monitoring continues; if a change in the data access mode is detected, the intelligent prefetching algorithm module and the cross-level data migration module are notified. After further receiving the notification of the change in the data access mode through the intelligent prefetching algorithm module, the processing starts from the input feature definition. The input features include historical data features, current state features and context features, which are combined into an input feature vector. Use function g to predict the cache block based on the input feature vector The miss rate of the cache block is calculated through a series of processing, and the predicted access frequency is finally calculated through the weight matrix and bias terms. If it is predicted that some data may be frequently accessed, the data will be pre-fetched into the high-level cache in advance. After receiving the notification of the change in data access pattern, the cross-level data migration module of the system also starts from the input feature definition and shares the input features with the intelligent pre-fetch algorithm module. The miss rate of the cache block is predicted based on the input features. The penalty cost of the miss is determined according to the cache block location, including storage cost and energy consumption cost, and then the total cost is calculated. If the total cost calculation result shows that it is cost-effective to migrate the data from the slower level of memory to the faster level of cache, the data migration operation is performed. The system continuously monitors the data access pattern, and each module continuously adjusts and optimizes the storage location of the data according to the changes to improve system performance and response speed.

[0100] It can be seen that an embodiment of the present invention provides a cross-level data pre-fetching method, which is applied to a preset data allocation and management system, including: after a business activity is updated, real-time monitoring of whether the data access mode related to the business activity has changed; if the data access mode has changed, using the target data access frequency prediction model of the preset data allocation and management system to predict the access frequency under the data access mode, and obtain a corresponding access frequency prediction result; wherein the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target correspondence relationship, and the target correspondence relationship is the correspondence between each system state of the preset data allocation and management system under different data access modes, context information related to data access under different data access modes, and the corresponding time information when executing different data access modes and the access frequency of the accessed data under different data access modes; according to the access frequency prediction result, the target data located in the memory that needs to be accessed within the next access time is predicted, and the predicted target data is pre-migrated from the memory to the preset cache, so as to use the preset cache to respond to the data access request initiated for the target data within the next access time.

[0101] It can be seen from the above technical solutions that by real-time monitoring of changes in data access patterns, the dynamic adjustment of data access behavior can be captured in a timely manner, no longer limited to the fixed patterns of traditional strategies, and thus better adapted to complex and changeable data access situations. Utilizing the target data access frequency prediction model, the access frequency is predicted by comprehensively considering multiple factors such as system status, context information, and time information. Compared with traditional strategies, it can more accurately predict data requirements, thereby improving the matching degree between data in the cache and actual access requirements, and significantly improving the cache hit rate. Determine the target access data based on accurate prediction results and migrate it to the internal memory. When the system initiates an access request, it can extract data from the fast internal memory, greatly reducing data access delays, significantly improving system performance and response speed, and improving user experience. The system performance is optimized and the response speed is improved.

[0102] Reference Figure 4 As shown, the present invention also provides a cross-level data pre-fetching device, which is applied to a preset data allocation management system, including:

[0103] A monitoring module 11 is used to monitor in real time whether a data access mode related to a business activity changes after the business activity is updated;

[0104] A prediction module 12 is used to predict the access frequency under the data access mode using the target data access frequency prediction model of the preset data allocation and management system if the data access mode changes, and obtain a corresponding access frequency prediction result; wherein the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target corresponding relationship, and the target corresponding relationship is the corresponding relationship between each system state of the preset data allocation and management system under different data access modes, context information related to data access under different data access modes, time information corresponding to the execution of different data access modes, and the access frequency of the accessed data under different data access modes;

[0105] The data pre-fetch module 13 is used to predict the target data located in the memory that needs to be accessed within the next access time based on the access frequency prediction result, and pre-migrate the predicted target data from the memory to a preset cache, so as to use the preset cache to respond to the data access request initiated for the target data within the next access time.

[0106] It can be seen that the embodiment of the present invention provides real-time monitoring of whether the data access mode related to the business activity has changed after the business activity is updated; if the data access mode has changed, the target data access frequency prediction model of the preset data allocation management system is used to predict the access frequency under the data access mode to obtain the corresponding access frequency prediction result; wherein, the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target correspondence relationship, and the target correspondence relationship is the correspondence between the system states of the preset data allocation management system under different data access modes, the context information related to data access under different data access modes, the corresponding time information when executing different data access modes, and the access frequency of the accessed data under different data access modes; according to the access frequency prediction result, the target data located in the memory that needs to be accessed within the next access time is predicted, and the predicted target data is pre-migrated from the memory to the preset cache, so as to use the preset cache to respond to the data access request initiated for the target data within the next access time. It can be seen that by monitoring the changes in the data access mode in real time, the dynamic adjustment of data access behavior can be captured in time, and it is no longer limited to the fixed mode of the traditional strategy, so as to better adapt to the complex and changeable data access situation. By using the target data access frequency prediction model, the access frequency is predicted by comprehensively considering multiple factors such as system status, context information, and time information. Compared with traditional strategies, it can more accurately predict data requirements, thereby improving the matching degree between cache data and actual access requirements, and significantly improving cache hit rate. According to the accurate prediction results, the target access data is determined and migrated to the internal memory. When the system initiates an access request, the data can be extracted from the fast internal memory, which greatly reduces data access delay, significantly improves system performance and response speed, and improves user experience. It achieves the optimization of system performance and the improvement of response speed.

[0107] Furthermore, the present application also discloses an electronic device. Figure 5 This is a structural diagram of an electronic device according to an exemplary embodiment, and the content in the diagram cannot be regarded as any limitation on the scope of use of this application. The electronic device may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the cross-level data prefetching method disclosed in any of the aforementioned embodiments. In addition, the electronic device in this embodiment may specifically be an electronic computer.

[0108] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0109] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0110] The operating system 221 is used to manage and control various hardware devices on the electronic device and the computer program 222, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the cross-level data prefetching method performed by the electronic device disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0111] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the cross-level data prefetching method disclosed above. For the specific steps of the method, reference may be made to the corresponding contents disclosed in the above embodiments, and no further description will be given here.

[0112] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0113] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0114] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0115] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0116] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A cross-level data pre-fetching method, characterized in that: Applied to preset data deployment management system, including: After a business activity is updated, monitor in real time whether the data access pattern related to the business activity has changed; If the data access mode changes, the target data access frequency prediction model of the preset data allocation and management system is used to predict the access frequency under the data access mode to obtain a corresponding access frequency prediction result; wherein, the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target corresponding relationship, and the target corresponding relationship is the corresponding relationship between each system state of the preset data allocation and management system under different data access modes, context information related to data access under different data access modes, time information corresponding to the execution of different data access modes, and the access frequency of the accessed data under different data access modes; Based on the access frequency prediction result, the target data located in the memory that needs to be accessed within the next access time is predicted, and the predicted target data is pre-migrated from the memory to a preset cache, so that the preset cache can be used to respond to the data access request initiated for the target data within the next access time.

2. The cross-level data pre-fetching method according to claim 1, characterized in that: Before predicting the access frequency under the data access mode using the target data access frequency prediction model of the preset data allocation management system, the method further includes: Collect the historical access frequencies of the corresponding historically accessed data under different data access modes of the preset data allocation management system, the historical system status of the preset data allocation management system under the corresponding data access mode, the historical context information related to the data access under the corresponding data access mode, and the historical time information corresponding to the execution of the corresponding data access mode, so as to obtain a data training set; The initial data access frequency prediction model is trained using the data training set to update the weight parameters of the initial data access frequency prediction model to obtain a target data access frequency prediction model; wherein the initial data access frequency prediction model is an initial model constructed based on a system state calculation factor, a weight parameter of the system state calculation factor, an access frequency calculation factor of accessed data, a weight parameter of the access frequency calculation factor, a context feature factor, a weight parameter of the context feature factor, a time feature factor, and a weight parameter of the time feature factor.

3. The cross-level data pre-fetching method according to claim 2, characterized in that: The using the data training set to perform model training on the initial data access frequency prediction model to update each weight parameter of the initial data access frequency prediction model to obtain a target data access frequency prediction model includes: The data training set is input into an initial data access frequency prediction model including an input layer, a long short-term memory neural network layer, a fully connected layer and an output layer connected in sequence, so that the historical feature vector obtained after the data training set is processed by the input layer is processed by the long short-term memory neural network layer of the initial data access frequency prediction model according to the cell state update rule and the hidden state update rule, so as to adjust and obtain the corresponding hidden state according to the historical time information, and then the hidden state is mapped in a high-dimensional manner through the fully connected layer to obtain the mapped data, and the mapped data is adjusted in importance and globally adjusted through the output layer to output the predicted access frequency of the historical accessed data; The weight parameters in the initial data access frequency prediction model are adjusted according to the predicted access frequency and the corresponding historical access frequency of the historically accessed data to obtain a target data access frequency prediction model.

4. The cross-level data pre-fetching method according to claim 3, characterized in that: Before the long short-term memory neural network layer of the initial data access frequency prediction model processes the historical feature vector obtained after the data training set is processed by the input layer according to the cell state update rule and the hidden state update rule, it also includes: The data training set is input into the input layer of the initial data access frequency prediction model so that the input layer constructs a historical feature vector under the corresponding data access mode based on the historical system status under different data access modes in the data training set, the historical context information related to data access under the corresponding data access mode, and the historical access frequency of the corresponding historical accessed data when the corresponding data access mode is executed.

5. The cross-level data pre-fetching method according to any one of claims 1 to 4, characterized in that: The step of pre-migrating the predicted target data from the memory to a preset cache includes: Acquire a first priority storage strategy of the target data at the current moment; the first priority storage strategy is a storage strategy of a target storage space in the memory storing the target data; Execute the step of updating the first priority storage strategy of the target data to a second priority storage strategy, and obtain the second priority storage strategy of the target data, so as to pre-migrate the target data from the target storage space of the memory to the first-level cache of the central processing unit; Among them, the second priority storage strategy is a storage strategy for storing the target data in the first-level cache of the central processing unit; the data access speed in the second priority storage strategy is higher than the data access speed in the first priority storage strategy.

6. The cross-level data pre-fetching method according to claim 5, characterized in that: Before obtaining the first priority storage strategy of the target data at the current moment, the method further includes: If the number of data accesses to each accessed data under the historical access time is less than a preset access number threshold, a first priority storage strategy is constructed to characterize the storage space of each accessed data stored in the memory under the historical access time according to the data heat, data access frequency, data freshness and system load of each accessed data and the corresponding parameter coefficients; wherein the data freshness is the inverse of the historical access time of the accessed data; Migrating each of the accessed data to a corresponding storage space in the memory according to the first priority storage strategy, generating a data cache strategy for each storage space in the memory and setting a corresponding data cache size; If the number of data accesses for each accessed data at the historical access time is greater than or equal to a preset access number threshold, a second priority storage strategy is constructed according to the data heat, data access frequency, data freshness and system load of each accessed data and the corresponding parameter coefficients thereof to store each accessed data at the historical access time in the cache of the central processing unit; When receiving an access request for the accessed data, the accessed data is extracted from the corresponding storage location through the preset data allocation management system; wherein the data cache size and data cache strategy of the cache and the storage space are adjusted according to each of the access frequency prediction results.

7. The cross-level data pre-fetching method according to claim 6, characterized in that: Also includes: Obtaining the storage cost of the memory, the storage cost in the central processor, the system load cost of the preset data allocation management system occupied by the access frequency prediction results of each accessed data predicted by the target data access frequency prediction model, and the time cost generated by obtaining the accessed data from the storage location, so as to obtain the cache strategy adjustment factor; The cache policy adjustment factor is used to adjust the storage space of the memory and the data cache size and data cache policy of the cache of the central processing unit.

8. A cross-layer data pre-fetching device, characterized in that: Applied to preset data deployment management system, including: A monitoring module, used to monitor in real time whether a data access mode related to a business activity changes after the business activity is updated; A prediction module, which is used to predict the access frequency under the data access mode using the target data access frequency prediction model of the preset data allocation and management system if the data access mode changes, and obtain a corresponding access frequency prediction result; wherein the target data access frequency prediction model is a model obtained after the initial data access frequency prediction model learns the target corresponding relationship, and the target corresponding relationship is the corresponding relationship between each system state of the preset data allocation and management system under different data access modes, the context information related to data access under different data access modes, the time information corresponding to the execution of different data access modes and the access frequency of the accessed data under different data access modes; A data prefetch module is used to predict the target data located in the memory that needs to be accessed within the next access time based on the access frequency prediction result, and to pre-migrate the predicted target data from the memory to a preset cache so that the preset cache can be used to respond to the data access request initiated for the target data within the next access time.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the cross-level data prefetching method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the cross-level data pre-fetching method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Cache management method, electronic device, storage medium and program product

    CN119620961A

  • Data access method, data access management system, storage medium and electronic equipment

    CN119828960A

  • Large model operation method and device based on edge device, equipment and medium

    CN120196449A

  • Self-adaptive dynamic priority caching method, system, equipment and medium

    CN120578349A

  • Cache space adjustment method and electronic equipment

    CN120743968A