Intent-aware data pre-fetching method, device and storage medium for solid-state storage device

Abstract

The present invention relates to the field of computer storage technology, and specifically to an intent-aware data pre-fetching method, device and storage medium for solid-state storage devices. The method comprises the following steps: collecting user behavior data, the user behavior data includes application usage frequency, file access time, system call log, hard disk read track and network traffic information; classifying, filtering and pre-processing the collected user behavior data, generating time series and behavior series, and mapping the time series and behavior series into multi-dimensional feature vectors; the present invention proposes an intent-aware data pre-fetching method for solid-state storage devices, combined with real-time analysis of user operation intentions, to achieve dynamic and efficient data pre-fetching, thereby improving the data access speed and overall performance of solid-state storage devices.

Description

Technical Field

[0001] The present invention relates to the field of computer storage technology, and in particular to an intent-aware data pre-fetching method, device and storage medium for a solid-state storage device. Background Art

[0002] Solid-state storage devices (such as SSDs and embedded storage) have been widely used in various electronic products due to their high read and write speeds, low power consumption, and durability. However, with the increasing complexity of applications and the amount of data, existing solid-state storage devices face severe challenges in data access speed and efficiency. Data prefetching is expected to break through this bottleneck.

[0003] In the prior art, most data prefetching methods rely on static prefetching strategies, which usually prefetch the data that may be needed by analyzing historical data access patterns, such as sequential reading or frequent access to fixed blocks, and cannot be intelligently adjusted according to the dynamic changes in user operations. However, the effectiveness of these static strategies is insufficient when faced with complex and changeable user operation scenarios. Therefore, the technical limitations of the existing methods are: static prefetching strategies cannot adapt to highly variable operating environments, cannot perceive user operation intentions in real time, resulting in a mismatch between prefetched data and actual needs, thereby failing to effectively improve the speed of data access. Summary of the invention

[0004] The present invention provides an intention-aware data pre-fetching method, device and storage medium for a solid-state storage device to perceive user operation intentions in real time, dynamically adjust pre-fetching strategies, and improve data access speed.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] Provided are a method, apparatus and storage medium for pre-fetching intent-aware data of a solid-state storage device, the method comprising the following steps:

[0007] Collect user behavior data, including application usage frequency, file access time, system call logs, hard disk read traces, and network traffic information;

[0008] Classify, filter, and preprocess the collected user behavior data to generate time series and behavior sequences, and map the time series and behavior sequences into multi-dimensional feature vectors;

[0009] Utilize multi-dimensional feature vectors and comprehensive reasoning decision tree algorithm to model user behavior and generate a behavior prediction model;

[0010] According to the behavior prediction model, the user's next operation is inferred and a prefetch instruction is generated. The prefetch instruction includes prefetch decision information and a priority list.

[0011] According to the prefetch instruction, the target data is prefetched from different storage layers of the solid-state storage device and stored in the designated cache area according to the priority;

[0012] Flexible cache size adjustment and cache replacement are used to dynamically manage the target data in the cache area in real time.

[0013] In another aspect, an apparatus for pre-fetching data with intent-awareness of a solid-state storage device is provided, the apparatus comprising:

[0014] The data collection module is used to collect application usage frequency, file access time, system call logs, hard disk read traces and network traffic information through a multi-source data collection mechanism;

[0015] The multi-dimensional behavior analysis module processes and analyzes the data obtained from the data acquisition module, generates time series and behavior sequences, and maps the time series and behavior sequences into multi-dimensional feature vectors;

[0016] User behavior modeling module, based on multi-dimensional feature vectors, uses a comprehensive reasoning decision tree algorithm to build a user behavior prediction model;

[0017] The pre-fetch decision generation module uses the user behavior prediction model to infer the user's next operation and generate pre-fetch decision information and priority list;

[0018] A data pre-fetch and cache module, used to pre-fetch required data blocks from the solid-state storage device and store them in a cache area according to the pre-fetch decision information;

[0019] The cache and data management module is used to dynamically manage the cache area using elastic cache size adjustment and cache replacement.

[0020] On the other hand, a non-volatile readable storage medium is provided for implementing intent-aware data prefetching, wherein the storage medium stores computer-readable instructions, which, when executed by a computer, enable the computer to implement the above-mentioned intent-aware data prefetching method for a solid-state storage device.

[0021] The beneficial effects of the present invention are:

[0022] The present invention proposes an intention-aware data prefetching method for solid-state storage devices, combined with real-time analysis of user operation intentions, to achieve dynamic and efficient data prefetching, thereby improving the data access speed and overall performance of solid-state storage devices.

[0023] Specifically, by generating simulated time series data, including multiple dimensional attributes and user operation labels, the diversity and extensiveness of the data are ensured, and the complexity of real-world operation scenarios can be fully simulated. Secondly, encoding and standardization processing are used to encode categorical variables and standardize feature data, so that the input data can better adapt to the analysis model, solving the problem that traditional methods cannot accurately reflect user operation intentions. In addition, by training the decision tree model, the accuracy and generalization ability of operation intention recognition are improved, ensuring that the pre-fetched data is highly matched with actual needs, and maximizing the pre-fetching efficiency. At the same time, by dynamically adjusting the data pre-fetching strategy, the system can respond to user operation changes in real time in a high-concurrency, multi-tasking environment, significantly reducing the latency of storage devices and improving the speed of data access.

[0024] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flow chart of an intent-aware data prefetching method according to an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of a decision tree behavior prediction model according to an embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a prefetch instruction in an embodiment of the present invention;

[0028] Figure 4 A flowchart of collecting user behavior data in one embodiment of the present invention;

[0029] Figure 5 A flowchart of mapping into a multi-dimensional feature vector in one embodiment of the present invention;

[0030] Figure 6 A flowchart of storing data in a designated buffer area according to priority in one embodiment of the present invention;

[0031] Figure 7 The figure is a flow chart of user feedback correction in one embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention provides the following preferred embodiments:

[0033] Embodiment 1

[0034] In order to solve the data access speed and efficiency problems of existing solid-state storage devices in high concurrency and multi-tasking environments, this embodiment proposes an intent-aware data prefetching method for solid-state storage devices. By integrating user behavior data collection and data analysis technology, the prefetching decision is made dynamic, thereby improving the speed of data access.

[0035] like Figure 1 As shown, the method comprises the following steps:

[0036] S101. Collect user behavior data, which includes application usage frequency, file access time, system call log, hard disk read track and network traffic information.

[0037] S102: Classify, filter, and preprocess the collected user behavior data to generate time series and behavior series, and map the time series and behavior series into multi-dimensional feature vectors.

[0038] S103, using multi-dimensional feature vectors, modeling user behavior through a comprehensive reasoning decision tree algorithm to generate a behavior prediction model.

[0039] S104: infer the user's next operation based on the behavior prediction model and generate a prefetch instruction, which includes prefetch decision information and a priority list.

[0040] S105 . Pre-fetch target data from different storage layers of the solid-state storage device according to the pre-fetch instruction, and store the data in a designated cache area according to priority.

[0041] S106: Use elastic cache size adjustment and cache replacement to dynamically manage the target data in the cache area in real time.

[0042] Specifically, the collection of user behavior data includes information in multiple dimensions, including application usage frequency, file access time, system call logs, hard disk read traces, and network traffic information. These behavior data are recorded in real time via sensors and monitoring modules to ensure the comprehensiveness and real-time nature of the data. It should be understood that this comprehensive collection of multi-source data can provide a comprehensive view of user operations and provide a basis for subsequent data analysis and pre-fetching decisions.

[0043] Furthermore, the collected user behavior data is classified, filtered and preprocessed in order to eliminate noise and redundant data, improve data quality and analysis efficiency. Specifically, the classification step includes dividing the data according to different operation types and sources, such as listing file access time and system call logs as independent categories, and combining hard disk read traces and network traffic information for multi-level analysis. The filtering step is to remove abnormal data points and duplicate records by setting thresholds and rules. The preprocessing step mainly includes data cleaning and data format conversion, converting the original data into standardized time series and behavior sequences for subsequent analysis and processing. These time series and behavior sequences are converted into multi-dimensional feature vectors through mapping rules, which can more accurately represent the characteristics of user behavior.

[0044] Furthermore, the generated multi-dimensional feature vectors are used to model user behavior through a comprehensive inference decision tree algorithm, aiming to build a behavior prediction model that can accurately predict the user's next operation, such as Figure 2 As shown. It is important to understand that the comprehensive reasoning decision tree algorithm combines the characteristics of the decision tree model such as easy explanation and high precision in this process. By selecting and splitting features of different dimensions, a hierarchical decision path is generated to explore the potential laws and patterns of user behavior. Through training and learning of historical user behavior data, the generated behavior prediction model can achieve accurate prediction of future operations.

[0045] Furthermore, based on the behavior prediction model, the user's next operation is inferred and a prefetch instruction is generated, such as Figure 3 As shown in the figure. These prefetch instructions contain not only prefetch decision information, but also a priority list to guide specific data prefetch operations. The prefetch decision information indicates the target data to be prefetched, while the priority list sorts the target data according to the urgency and possibility of user operations, providing a basis for setting the priority of prefetch operations. This priority mechanism can ensure that the data most likely to be accessed is prefetched first during the prefetch process, thereby improving the overall efficiency of the storage system.

[0046] Further, according to the prefetch instruction, the target data is prefetched from different storage layers of the solid-state storage device and stored in the designated cache area according to priority. In the present embodiment, the solid-state storage device generally includes multiple storage layers, such as a cache layer, a main storage layer, and a cold storage layer. High-priority data will be prefetched from the cache layer first, while low-priority data may be prefetched from the main storage layer or the cold storage layer, and finally stored in the cache area according to priority to ensure efficient storage resource utilization and access speed. It should be understood that the operation of storing in the cache area according to priority is intended to optimize the cache hit rate and data accessibility.

[0047] The benefit of this embodiment is that the target data in the cache area is managed dynamically in real time through elastic cache size adjustment and cache replacement strategy. The elastic cache size adjustment strategy dynamically adjusts the cache size according to the current system load and task requirements to adapt to different operating environments and demand conditions. The cache replacement strategy ensures that the data in the cache is always the latest and most needed by setting intelligent replacement rules, such as based on least used (LRU) or other advanced algorithms, thereby improving the utilization efficiency of the cache space.

[0048] Through this embodiment, accurate perception of user intentions and dynamic data pre-fetching are achieved, and the data access strategy of the solid-state storage device is optimized at multiple levels, so that the system can always provide users with fast response and efficient storage services. This comprehensive optimization strategy not only enhances the accuracy of data pre-fetching, but also improves the performance of solid-state storage devices in practical applications.

[0049] Embodiment 2

[0050] In order to solve the performance bottleneck problem of existing solid-state storage devices in high concurrency and multi-tasking environments, this embodiment further refines the steps of user behavior data collection, aiming to achieve more accurate intent perception and data pre-fetching through more comprehensive and detailed user operation records.

[0051] like Figure 4 As shown, the steps of collecting user behavior data include:

[0052] S201. Capture and record the user's operation records in different applications, including opening applications, editing files, saving and closing.

[0053] S202. Record access logs in the file system, including file creation, modification, deletion, and read and write operations.

[0054] S203: Monitor and record the user's network activities, including traffic volume, accessed IP address, and access time.

[0055] It is understandable that the operation records of users in different applications are the basic conditions for collecting user behavior data. These records include detailed data generated by users when opening applications, editing files, saving and closing them. It is important to understand that these operation records are captured by embedded recording systems or application monitoring modules and uploaded to the data analysis center in real time. By timestamping and sorting these operation records, the system can reconstruct the user's operation path and behavior pattern, thereby providing accurate data input for the behavior prediction model.

[0056] Furthermore, the access log in the file system is another data collection source. The access log records the creation, modification, deletion, and read and write operations of the file in detail. These log information include file path, operation type, timestamp, and operation user. It is understandable that by classifying and analyzing these access logs, the system can identify the user's behavior habits and access frequency in the file system, thereby inferring possible future operations. For example, frequently edited and saved files may be accessed multiple times in a short period of time in the future. The system can prioritize pre-fetching these files to increase data reading speed.

[0057] Furthermore, the monitoring and recording of users' network activities are also the content of data pre-fetching strategy optimization. Specifically, network activity data includes information such as traffic size, access IP address and access time. These data are captured in real time by the network monitoring module and recorded in the behavior log. It can be understood that these network activity information not only reflects the user's online behavior pattern, but also can infer the user's interest points and operation trends through the correlation analysis of IP address and access time. For example, data that frequently accesses a remote server can be pre-fetched into the local cache to reduce the access time cost caused by network delay.

[0058] Furthermore, when the above-mentioned multi-source data is processed comprehensively, this embodiment adopts a series of data preprocessing and analysis technologies. First, through data classification and filtering steps, user operation records, file system access logs and network activity data are separated and cleaned according to different dimensions to remove noise and invalid data. Then, the preprocessed data is converted into standardized time series and behavior sequences for further mapping to generate multidimensional feature vectors. It should be understood that the multidimensional feature vector is the basis of the subsequent comprehensive reasoning decision tree algorithm. Through the training and learning of these feature vectors, the behavior prediction model can more accurately reflect the user's operational intentions.

[0059] Furthermore, by comprehensively utilizing user operation records, file system access logs, and network activity data, this embodiment constructs a user behavior data collection and analysis system. This system not only improves the accuracy of the behavior model and the precision of the pre-fetching strategy through multi-dimensional data sources, but also enhances the adaptability of the system in different application scenarios. The pre-fetch instructions generated by the behavior prediction model accurately predict the next operation based on the user's operating rules and behavior patterns, and sort them through a priority list, thereby guiding the pre-fetching and cache management of data. The priority list not only includes the importance ranking of the pre-fetched data, but also combines multiple factors such as data access frequency and timeliness to ensure that the pre-fetched data can maximize the system response speed.

[0060] Furthermore, in the process of implementing data prefetching, this embodiment pays attention to the management strategy of the cache area in real time, and dynamically adjusts the cache size according to the current system load and data access frequency through the elastic cache size adjustment mechanism, so that the cache resources can be optimally configured under different load conditions. At the same time, a cache replacement mechanism is adopted to ensure that the data in the cache is always the latest and most valuable by real-time monitoring and analysis of the access to cache data, using, for example, a least used (LRU) or frequency-based replacement algorithm, thereby improving cache utilization and system performance.

[0061] The benefit of this embodiment is that by refining the collection and analysis of user behavior data, combined with multi-dimensional information such as application operation records, file system access logs and network activity data, a more accurate and efficient intent-aware data pre-fetching method is constructed. Through the behavior prediction model and intelligent cache management strategy, dynamic optimization of data pre-fetching and efficient use of cache resources are achieved, and the data access speed of solid-state storage devices is improved.

[0062] Embodiment 3

[0063] In order to address the shortcomings of existing solid-state storage devices in data access speed and pre-fetching efficiency, this embodiment further refines the processing flow of user behavior data, classifies, filters and pre-processes the collected user behavior data to generate time series and behavior sequences, and maps them into multi-dimensional feature vectors, providing high-quality input data to support intelligent analysis and pre-fetching decisions.

[0064] like Figure 5 As shown, the steps of classifying, filtering, and preprocessing the collected user behavior data, generating time series and behavior series, and mapping them into multi-dimensional feature vectors include:

[0065] S301. Convert user behavior data into a unified format and filter out noise and outliers.

[0066] S302: extracting features from the processed user behavior data to obtain feature data, where the feature data includes time features, frequency features, data type features, user interaction mode features, and network access features.

[0067] S303: Assemble the feature data into a multi-dimensional feature vector and perform vectorization processing according to specific requirements.

[0068] Specifically, the user behavior data is first converted into a unified format to implement the data preprocessing step. What needs to be understood in this process is that user behavior data comes from various sources, including application logs, system calls, file operations, and network traffic, etc. These data formats are different and may contain a lot of noise and outliers. In order to eliminate these interferences, the system first standardizes the original data and converts data in different formats into a unified structured format, such as unifying timestamps into a unified time format, converting string type data into numeric types, etc. Then, the system uses filtering algorithms to remove noise and outliers to ensure the accuracy and reliability of the data. Using anomaly detection algorithms in machine learning, such as isolation forests or statistical-based outlier detection, abnormal data can be effectively identified and filtered out to improve data quality.

[0069] Furthermore, feature extraction is performed on the processed user behavior data to convert the behavior data into analyzable feature data. During the feature extraction process, the system analyzes and calculates from multiple dimensions such as time features, frequency features, data type features, user interaction pattern features, and network access features. For example, time features can include the specific time point and time interval of the operation; frequency features can include the number of repetitions and change trends of a certain behavior within a certain time window; data type features include attributes such as file type and operation type; user interaction pattern features focus on the behavior patterns exhibited by users during the operation process, such as continuous clicks, drag operations, etc.; network access features cover the frequency of network requests, data transmission volume, and access targets. These features are converted into quantitative feature data through feature engineering methods such as standardization, normalization, and principal component analysis (PCA), laying the foundation for subsequent vectorization processing.

[0070] Furthermore, the extracted feature data is assembled and vectorized to form a multi-dimensional feature vector. In this process, the system first assembles various types of feature data in a predefined format, including data combinations of multiple dimensions such as time features and frequency features. Then, these feature data are vectorized according to specific needs. For example, each feature data can be converted into a fixed-length feature vector through a specific mapping function, where each dimension represents the value of a behavioral feature. What needs to be understood in this process is that vectorization processing not only helps to improve the computability of behavioral data, but also reduces computational complexity and storage occupancy by reducing feature dimensions.

[0071] Furthermore, in the implementation process, this embodiment significantly improves the extraction speed and accuracy of feature data by selecting efficient feature extraction algorithms and vectorization methods. It is understandable that different types of user behavior data may require different feature extraction methods to achieve the best results. For example, for time series data, the use of time windows and sliding average methods can effectively capture time-varying features; for frequency features, signal processing methods such as Fourier transform and wavelet analysis can be used; for interactive mode features and network access features, user behavior pattern analysis and network traffic analysis methods can be combined to extract more meaningful features.

[0072] Through this embodiment, a systematic user behavior data processing and feature extraction process is constructed, from the initial capture of data, format conversion, outlier filtering, to the extraction and vectorization of multi-dimensional features, to achieve high-quality, multi-dimensional feature vector generation. These feature vectors provide accurate input data for the subsequent comprehensive reasoning decision tree algorithm, improve the accuracy of the behavior prediction model and the rationality of the pre-fetching decision. By optimizing the data processing and analysis process, this embodiment not only improves the performance of solid-state storage devices in terms of data access speed and pre-fetching efficiency, but also provides a solid technical foundation for realizing dynamic data pre-fetching strategies.

[0073] Embodiment 4

[0074] In order to solve the problem of inaccurate models in existing solid-state storage devices when predicting user behavior and pre-fetching data accordingly, this embodiment further refines the process of generating a behavior prediction model using a multi-dimensional feature vector, and constructs a more accurate user behavior model through a comprehensive reasoning decision tree algorithm, thereby improving the efficiency of data pre-fetching.

[0075] Furthermore, in this embodiment, the multidimensional feature vector extracted from the user behavior data is first converted into a behavior data matrix. It should be understood that this behavior data matrix integrates the user's operation characteristics in different situations and reflects the overall picture of his historical behavior. By analyzing the user's personal historical behavior, a behavior feature matrix is ​​generated. This matrix is ​​the basis for behavior prediction and contains important information such as the frequency, time interval, and operation type of the user's behavior.

[0076] Furthermore, based on the generated behavior feature matrix, this embodiment constructs the root node of the comprehensive reasoning decision tree. The root node is not just a simple starting point, it contains the probability of behavior occurrence and the distribution of potential behaviors, representing the most likely operation of the user in the current situation. The construction of the root node is based on the analysis and induction of a large amount of historical behavior data, covering the common behavior patterns and operation tendencies of users in similar situations. It can be understood that the probability of behavior occurrence contained in the root node is obtained through statistical analysis and plays a decisive guiding role in the entire reasoning process.

[0077] Furthermore, by gradually expanding and splitting the root node, the user's possible subsequent behavior is inferred, and finally a behavior prediction model is generated. In this process, the comprehensive reasoning decision tree algorithm continuously splits the feature vector and analyzes the user's behavior characteristics layer by layer, thereby constructing a detailed behavior path. Each root node represents a possible behavior choice. After multiple splits, the branches of the decision tree are gradually enriched until a complete behavior prediction model is generated. It can be understood that this model not only includes the prediction of the user's current operation, but also can infer a series of operations of the user in subsequent steps, providing a reliable basis for the formulation of data pre-fetching strategies.

[0078] Furthermore, in order to improve the adaptability of the model, this embodiment adopts an adaptive weight adjustment strategy. This strategy adjusts the weights of each behavior branch to adapt to the dynamic changes of user behavior. Specifically, the calculation formula for weight adjustment is:

[0079]

[0080] Among them, α i,new is the updated weight, α i is the initial weight, ζ is the adjustment factor, ΔP(B i |F) is the change in the probability of the behavior occurring, and N is the number of root nodes.

[0081] When the user’s actual behavior deviates from the prediction, the weight distribution of the model is quickly adjusted by increasing ζ, so that the model can adapt to the changing user behavior more quickly, thereby improving the accuracy of the prediction.

[0082] Through the improvement of this embodiment, a dynamic behavior prediction model that can adapt to changes in user behavior in real time is constructed, which significantly improves the accuracy and efficiency of data pre-fetching. By comprehensively using the behavior data matrix, feature matrix and decision tree algorithm, and combining the adaptive weight adjustment strategy, the system can always maintain efficient pre-fetching performance in a complex and changeable user operation environment.

[0083] Embodiment 5

[0084] In order to solve the problems of low resource utilization and inaccurate prefetch instruction priority in the data prefetching process of existing solid-state storage devices, this embodiment further refines the priority calculation method of prefetching data blocks, and generates more accurate prefetch instructions by introducing multi-dimensional factors such as time, access frequency and data block size, combined with weight coefficients, thereby optimizing the efficiency of data prefetching.

[0085] Furthermore, in the calculation process of data pre-fetching priority, this embodiment first introduces the consideration of time dimension. It should be understood that the interval between the last access time of a data block and the current time directly affects whether the data block is likely to be accessed again in the future. If the time since a data block was last accessed is short, it is more likely to be accessed again in a short time; conversely, if the time since the last access is long, the necessity of pre-fetching the data block is relatively low. Therefore, by calculating the current time T now The last access time of the data block is T last (D i ) and compare it with the total time unit T total Divide by , and get the time priority factor of the data block. It can be understood that this factor reflects the pre-fetching demand of the data block in the time dimension. It can be understood that the weight coefficient δ1 is used to control the proportion of this factor in the final priority score.

[0086] Furthermore, this embodiment uses the access frequency of data blocks as another important priority evaluation criterion. Frequently accessed data blocks are usually key data in user operations, so they should be given higher priority in the prefetching process. The system monitors the access frequency F of each data block. access (D i ), and compare it with the total access frequency of all data blocks F total Compare and calculate the access frequency factor. It should be understood that by adding the access frequency factor, the system can effectively distinguish between high-frequency access data blocks and low-frequency access data blocks, thereby giving priority to pre-fetching core data that is frequently operated by users. It can be understood that the weight coefficient δ2 is used to adjust the influence of the access frequency factor in the total priority calculation, ensuring that data blocks with higher frequency can get higher priority scores.

[0087] Furthermore, the size S(D i ) is also one of the considerations for priority calculation in this embodiment. In a limited cache space, larger data blocks occupy more cache resources, and prefetching these data blocks may have a greater impact on the utilization of the cache space. Therefore, it is necessary to balance the relationship between the size of the data block and its prefetch priority. By calculating the size S(D i ) and the total data block size S totalBy calculating the ratio between the two and assigning a weight coefficient δ3, the system can appropriately lower its priority when prefetching a larger data block to ensure the effective use of the cache space. At the same time, it is also understandable that for a smaller data block, the system may increase its priority so that it can be prefetched more quickly to improve the response speed of the system.

[0088] Through the comprehensive calculation of the above time priority factor, access frequency factor and data block size factor, the priority score P(D i ). The specific priority calculation formula is:

[0089]

[0090] Among them, δ1, δ2, and δ3 are adjustable weight coefficients used to balance the impact of time, frequency, and size factors in priority scoring. It is understandable that different application scenarios and user behavior patterns may require different weight combinations to meet specific performance requirements. In actual applications, the system can dynamically adjust these three weight coefficients based on real-time workload and cache space conditions to adapt to changes in user behavior.

[0091] Furthermore, after the priority score is generated, the system sorts the data blocks according to the score, and determines which data blocks should be pre-fetched first according to the predetermined cache strategy. Through the priority calculation method of this embodiment, the accuracy and timeliness of data pre-fetching are improved, especially in a high-load, multi-tasking operating environment, which can effectively reduce unnecessary data reading operations and improve the response speed of the system.

[0092] The benefit of this embodiment is that a flexible priority calculation system is constructed by comprehensively considering multi-dimensional factors such as the time, access frequency and size of the data block, and combining the weight adjustment strategy. Through this system, the system can manage cache resources more intelligently, improve the efficiency of data prefetching, and maintain high stability performance in complex user operation environments.

[0093] Embodiment 6

[0094] In order to solve the problems of cache overflow and data transmission congestion that may be caused when prefetching target data in a solid-state storage device, this embodiment further refines the relevant features of data prefetching and cache management to optimize system performance and ensure fast access to high-priority data.

[0095] like Figure 6 As shown, according to the pre-fetch instruction, the steps of pre-fetching target data from different storage layers of the solid-state storage device and storing them in the designated cache area according to priority include:

[0096] S601. Based on the priority list in the prefetch instruction, selectively read a target data block from a solid-state storage device for data block location.

[0097] S602: Transmit the selected target data blocks to the cache area in segments as needed to prevent cache overflow and data transmission congestion.

[0098] S603 . Dynamically adjust the data storage structure in the cache area according to the priority list in the prefetch instruction to ensure that high-priority data can be quickly accessed.

[0099] It should be understood that, in the process of executing data prefetching, this embodiment selectively reads target data blocks from different storage layers of the solid-state storage device based on the priority list in the prefetch instruction. In actual operation, the solid-state storage device usually includes multiple layers of storage media, such as a cache layer, a main storage layer, etc., and the data access speeds and capacities of these storage layers are different. Therefore, the setting of the priority list determines which data blocks should be read from the storage layer and stored in the cache first.

[0100] Furthermore, after the data block is located, this embodiment adopts a data processing strategy of on-demand segmented transmission. The role of this strategy is to transmit the selected target data block to the cache area in segments on demand. It should be understood that on-demand segmented transmission can not only prevent the cache area from overflowing due to excessive data transmission, but also effectively avoid congestion during the data transmission process. This strategy is particularly suitable for scenarios with high concurrent data access, ensuring the stable operation of the system under high load conditions.

[0101] Furthermore, according to the priority list in the prefetch instruction, the present embodiment also dynamically adjusts the data storage structure in the cache area. It is understandable that the management of the cache area should not only consider the read and write efficiency of the data, but also take into account the access priority of different data. Therefore, during the prefetch process, the system will store high-priority data blocks in the cache area first according to the real-time adjustment of the priority list, and ensure that they can be accessed in the shortest time. This dynamic adjustment mechanism significantly improves the response speed of high-priority tasks and optimizes the utilization of the cache area.

[0102] Furthermore, in order to improve the efficiency of data transmission and storage, the present embodiment also optimizes the management strategy of the cache area. During the data transmission process, the system will monitor the usage status of the cache area in real time, and flexibly adjust the allocation scheme of the cache area according to the priority and access frequency of the data. For example, for high-priority and frequently accessed data blocks, the system will reserve more cache space for them to reduce the data transmission delay caused by insufficient cache space.

[0103] The benefit of this embodiment is that it effectively solves the performance bottleneck problem that may be encountered when performing data pre-fetching and cache management in a solid-state storage device. By dynamically adjusting the structure of the cache area and the data transmission strategy, this embodiment can significantly improve the access speed of data, especially when processing high-priority tasks, and can ensure the rapid response and stable operation of the system. At the same time, the strategy of on-demand segmented transmission also greatly reduces the risk of cache overflow and data transmission blockage, thereby improving the overall performance of the solid-state storage device in complex application scenarios. Through the optimization of this embodiment, the solid-state storage device can perform data pre-fetching and cache management more efficiently when facing large-scale data access requests.

[0104] Embodiment 7

[0105] In order to solve the problems of unreasonable cache resource allocation and inflexible data replacement strategy faced by solid-state storage devices when performing data prefetching, this embodiment further optimizes the dynamic management strategy of the cache area, specifically, it makes a detailed design on the calculation method of cache replacement to ensure that the system can efficiently handle data prefetching requirements in a multi-tasking environment.

[0106] In this embodiment, in the step of real-time dynamic management of target data in the cache area by using elastic cache size adjustment and cache replacement, the calculation method of cache replacement is:

[0107] , where R(D i ) is the data block D i The replacement priority, λ, η and σ are weight factors, T recency (D i ) is the last access time of the data block, F frequency (D i ) is the access frequency, S(D i ) is the size of the data block.

[0108] It should be explained that the system uses a cache replacement calculation formula based on multi-factor weighting by real-time monitoring of the cache usage to more reasonably determine the replacement priority of each data block. Specifically, the formula comprehensively considers the last access time T of the data block. recency (D i ), access frequency F frequency (D i ) and the size of the data block S(D i ), and introduces weight factors λ, η and σ, so that the replacement priority R(D i ) can more accurately reflect its importance in the current task scenario.

[0109] It is important to understand that the last access time T recency(D i ) reflects the time sensitivity of the data block. The system multiplies its reciprocal value by the weight factor λ, thereby ensuring to a greater extent that the recently accessed data block is not replaced prematurely. This design is particularly suitable for frequently accessed data sets. On the other hand, the access frequency F frequency (D i ) is processed logarithmically and multiplied by the weight factor η, which effectively alleviates the excessive occupation of cache resources by extremely high-frequency access data, and also avoids the situation where low-frequency access data is frequently replaced. Finally, the data block size S (D i ) is multiplied by the weight factor σ to reflect its occupancy of the cache area. This factor makes the system more cautious when replacing larger data blocks to reduce the waste of cache space.

[0110] Furthermore, this embodiment also introduces an elastic cache size adjustment mechanism. In actual operation, the size of the cache area should not be fixed, but should be dynamically adjusted according to the current system load and data access mode. Specifically, the system will evaluate the utilization efficiency of the current cache area in real time based on the cache hit rate and data access frequency. When it is found that the utilization rate of the cache area is too high and the hit rate decreases, the system will automatically trigger the cache expansion operation; conversely, if it is found that there are a large number of data blocks in the cache area that have not been accessed for a long time, the cache space will be appropriately reduced to free up resources for other tasks. This elastic adjustment mechanism not only improves the utilization efficiency of cache resources, but also can flexibly respond to data pre-fetching requirements in different scenarios.

[0111] Furthermore, in the process of data pre-fetching, the present embodiment further refines the management strategy of data in the cache area. The system will dynamically adjust the storage structure of the data blocks in the cache area according to the priority list of the current task to ensure that the high-priority data blocks are always in an easily accessible location. It should be understood that dynamic adjustment is not just a simple storage location adjustment, but also includes the optimization of the pre-fetching order and transmission strategy of the data blocks. For example, for high-priority data blocks that are about to be accessed, the system will give priority to reading from the storage medium and transmitting to the cache area, while temporarily moving low-priority data blocks to suboptimal storage locations to ensure rapid response of high-priority tasks.

[0112] Through the optimization of this embodiment, the solid-state storage device can perform data pre-fetching and cache management more intelligently, especially in the scenario of multi-task parallel processing, which can effectively avoid the waste of cache resources and the bottleneck problem of data transmission. By comprehensively considering the impact of various factors on the priority of data block replacement and combining a flexible cache size adjustment strategy, this embodiment achieves more efficient cache management and faster system response speed.

[0113] Embodiment 8

[0114] In order to solve the problems of insufficient accuracy of prefetching decision models and rapid changes in user needs when existing solid-state storage devices perform data prefetching, this embodiment further optimizes the correction mechanism of the prefetching decision model. By introducing user feedback correction steps, the prefetching model is continuously adjusted and optimized to improve the accuracy of data prefetching and the system response speed.

[0115] like Figure 7 As shown, the method also includes user feedback correction, which performs feedback correction on the actual operation results of the user and adjusts the pre-fetching decision model to improve the pre-fetching accuracy. The user feedback correction step includes:

[0116] S801: Perform statistics and analysis on the hit situation of pre-fetched data and cache hit rate.

[0117] S802: According to user operation feedback, dynamically adjust user behavior characteristics and model parameters to optimize the user intention prediction model.

[0118] S803: Use incremental learning and / or online learning algorithms to continuously optimize and enhance the pre-fetch decision model.

[0119] In this embodiment, the introduction of the user feedback correction mechanism is mainly aimed at the analysis and adjustment of the pre-fetched data hit situation and the cache hit rate. First, the system will compare the pre-fetched data with the actual access data and count the hit situation of the pre-fetched data. The hit rate directly reflects the prediction accuracy of the pre-fetch model, while the cache hit rate reflects the effectiveness of the cache management strategy. Through the statistics and analysis of these two indicators, the system can promptly discover the deficiencies in the pre-fetch model and cache strategy, thereby providing data support for subsequent model adjustments.

[0120] Furthermore, based on the above statistical results, the system will dynamically adjust user behavior characteristics and model parameters. This process includes re-evaluating characteristics such as the frequency, type, and time of user operations, and dynamically adjusting the model parameter settings based on the changing trend of user behavior. For example, when a user's behavior pattern changes significantly, such as frequent access to a specific type of file over a period of time, the system will increase the pre-fetching priority of files of that type to better adapt to the user's operating habits. It should be understood that this dynamic adjustment is not just a simple update of the model parameters, but also includes re-modeling of user behavior characteristics to ensure that the model can accurately reflect the user's current needs.

[0121] Furthermore, in order to further improve the accuracy and adaptability of the pre-fetch decision model, this embodiment adopts incremental learning and online learning algorithms. The incremental learning algorithm enables the system to gradually optimize the model while continuously accumulating user operation data, without having to train the entire model from scratch. The online learning algorithm allows the system to update the model in real time during the user operation process to ensure that the model is always in the optimal state. The combination of these two algorithms enables the pre-fetch decision model to quickly respond to changes in user behavior, improve the hit rate of data pre-fetching and the efficiency of cache management.

[0122] The benefit of this embodiment is that by introducing user feedback correction and incremental learning and online learning algorithms, the solid-state storage device can more efficiently adjust the pre-fetch strategy and improve the overall performance of the system. The system continuously monitors and analyzes the pre-fetch data and cache hit rate, and dynamically adjusts the pre-fetch model parameters, so that the system can better adapt to the user's operating habits and significantly improve the speed and accuracy of data access.

[0123] Embodiment 9

[0124] In order to solve the problems of insufficient prediction accuracy of user operation intentions and low cache management efficiency of existing solid-state storage devices during data prefetching, this embodiment further optimizes the module design of the data prefetching device to achieve more intelligent and efficient prefetching operations and cache management.

[0125] In this embodiment, the data acquisition module is responsible for acquiring user behavior data through a multi-source data acquisition mechanism. These data include but are not limited to application usage frequency, file access time, system call logs, hard disk read traces, and network traffic information. In order to achieve comprehensive and accurate data collection, the module is equipped with a variety of sensors and monitoring tools, which can capture user operation behaviors in real time and record them as log data that can be analyzed by the system. It should be understood that the accuracy and comprehensiveness of data collection directly determine the effect of subsequent analysis and model building. Therefore, this embodiment particularly optimizes the details of data collection, including unified processing of data formats and filtering of redundant information.

[0126] Furthermore, after acquiring the data transmitted by the data acquisition module, the multi-dimensional behavior analysis module first pre-processes the data, including noise filtering, data format standardization and other operations to ensure the high quality of the data. Subsequently, the module analyzes the processed data, generates time series and behavior sequences, and maps them into multi-dimensional feature vectors. These feature vectors can accurately reflect the user's operation mode, covering multiple dimensions such as time features, frequency features, data type features, etc., providing rich information for subsequent user behavior modeling. It should be understood that the generation process of multi-dimensional feature vectors needs to be combined with specific algorithms and rules to ensure the accuracy and availability of feature extraction.

[0127] Furthermore, the user behavior modeling module constructs a user behavior prediction model based on the multi-dimensional feature vector generated above through a comprehensive reasoning decision tree algorithm. This model can predict the user's next operation based on historical data and current operations. The decision tree algorithm was selected because it is efficient and easy to explain when processing multi-dimensional and multi-category data. During the model training process, the system optimizes feature selection and model parameters through multiple iterations, making the prediction results more accurate and reliable.

[0128] Furthermore, the prefetch decision generation module uses the prediction model provided by the user behavior modeling module to generate prefetch decision information and a priority list. The prefetch decision information indicates the target data blocks that need to be prefetched from the storage medium, and the priority list sorts these data blocks according to the predicted urgency, thereby ensuring that the system can prioritize prefetching those data that are most likely to be accessed by users. The design purpose of this module is to improve the efficiency of data prefetching and avoid slow system response caused by data not being loaded in time.

[0129] Furthermore, the data pre-fetch and cache module is responsible for executing specific data pre-fetch operations. According to the instructions of the pre-fetch decision generation module, the module selectively pre-fetches target data blocks from different storage layers of the solid-state storage device and stores these data blocks in the cache area. In order to improve the efficiency of data transmission and storage, this embodiment optimizes the cache management strategy, including dynamically adjusting the size of the cache area and the cache replacement strategy.

[0130] Furthermore, the cache and data management module dynamically manages the cache area. This module combines elastic cache resizing and cache replacement strategies to ensure efficient use of the cache area. The cache resizing mechanism allows the system to dynamically expand or reduce the cache space according to the current task requirements and cache usage, while the cache replacement strategy selectively removes or retains data blocks in the cache based on preset rules to ensure fast access to high-priority data and reasonable use of storage space.

[0131] The benefit of this embodiment is that, through modular design and optimization, the solid-state storage device can perform data pre-fetching and cache management more intelligently. Through multi-source data collection and multi-dimensional behavior analysis, the system can accurately predict the user's operation needs and adjust the data pre-fetching strategy in real time to ensure the efficiency and response speed of the system. Through the flexible cache management mechanism, the system can dynamically adjust the use of cache space according to actual needs, thereby improving the utilization efficiency of cache resources.

[0132] Embodiment 10

[0133] In order to solve the problem that the data prefetching method in the existing solid-state storage device cannot accurately perceive the user intention and the prefetching operation efficiency is low, this embodiment further optimizes and refines the computer-readable instructions stored on the non-volatile readable storage medium, so that the computer can efficiently implement the intention-aware data prefetching method when executing these instructions.

[0134] The present embodiment relates to a non-volatile readable storage medium, which stores a set of computer-readable instructions, which, when loaded and executed on a computer device, enable the computer to implement a series of data pre-fetch operations. These instructions are composed of multiple modular code segments, each of which implements a specific functional step to ensure the smoothness and efficiency of the entire pre-fetch process.

[0135] Furthermore, the data collection module code segment is used to collect user operation data from multiple source data such as application usage frequency, file access time, system call log, hard disk read track and network traffic information. This operation obtains and dynamically updates relevant data in real time by calling the underlying interface of the system to ensure the timeliness and accuracy of data collection. It can be understood that this step realizes cross-platform data collection capabilities by calling the API provided by the operating system.

[0136] Furthermore, the multi-dimensional behavior analysis module code segment preprocesses and analyzes the collected data. The preprocessing step includes denoising, standardization, and feature extraction, and the analysis step converts the processed data into time series and behavior series, and maps them into multi-dimensional feature vectors. These feature vectors form a multi-dimensional representation of user operation behavior, which is the basis for subsequent user behavior prediction.

[0137] Furthermore, the user behavior modeling module code segment builds a user behavior prediction model based on the above feature vectors through a comprehensive reasoning decision tree algorithm. The model training process is divided into two stages: offline training and online calibration. The offline training stage uses historical operation data for initial model training, while the online calibration stage uses incremental learning and real-time updates to ensure the accuracy and adaptability of the model in a dynamic environment.

[0138] Furthermore, the pre-fetch decision generation module code segment uses the trained user behavior prediction model to generate pre-fetch decision information and a priority list. By calling the optimized prediction algorithm, the module can efficiently predict the user's next possible operation and generate corresponding data pre-fetch instructions. These instructions include a list of data blocks to be pre-fetched and corresponding priority information to guide the execution of data pre-fetch operations.

[0139] Furthermore, after receiving the prefetch decision information, the data prefetch and cache module code segment performs specific data prefetch operations. The program calls the underlying storage control interface to obtain the target data blocks from the solid-state storage device and caches these data blocks into the cache area. This module ensures the efficiency and robustness of data transmission through an optimized storage read algorithm.

[0140] Furthermore, the cache and data management module code segment is responsible for the dynamic management of the cache area, including elastic cache size adjustment and cache replacement strategy. This code segment dynamically adjusts the size of the cache area according to the real-time cache hit rate and system load by calling the built-in cache management algorithm, and selectively removes low-priority data blocks according to the preset replacement strategy (such as LRU, LFU) to ensure efficient use of cache resources.

[0141] Through this embodiment, the non-volatile readable storage medium realizes efficient intention-aware data prefetching operation in the computer system. Through modular instruction design and multi-level optimization, the system can accurately perceive the user's operation intention and intelligently perform data prefetching and cache management. Not only does it effectively improve the accuracy of data prefetching and the system response speed, but it also ensures the efficiency of the system when processing multi-tasking operations through a flexible cache management mechanism.

[0142] The above embodiments further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intent-aware data pre-fetching method for a solid-state storage device, characterized in that: The method comprises the following steps: Collect user behavior data, including application usage frequency, file access time, system call logs, hard disk read traces, and network traffic information; Classifying, filtering, and preprocessing the collected user behavior data to generate time series and behavior sequences, and mapping the time series and behavior sequences into multidimensional feature vectors; Utilizing the multidimensional feature vector, user behavior modeling is performed through a comprehensive reasoning decision tree algorithm to generate a behavior prediction model; Inferring the user's next operation according to the behavior prediction model and generating a prefetch instruction, wherein the prefetch instruction includes prefetch decision information and a priority list; Prefetch target data from different storage layers of the solid-state storage device according to the prefetch instruction, and store them in a designated cache area according to priority; Real-time dynamic management of the target data in the cache area by using elastic cache size adjustment and cache replacement; The step of using the multi-dimensional feature vector to model the user behavior through a comprehensive reasoning decision tree algorithm to generate a behavior prediction model includes: Convert the multidimensional feature vector into a behavior data matrix, and generate a behavior feature matrix based on the user's personal historical behavior; Based on the behavior feature matrix, construct a root node of a comprehensive reasoning decision tree, wherein the root node includes a behavior occurrence probability and a potential behavior distribution; By gradually expanding and splitting the root node, inferring the user's possible subsequent behavior, and generating the behavior prediction model; Adopting the adaptive weight adjustment strategy, the weight adjustment calculation formula is: , where α i,new is the updated weight, α i is the initial weight, ζ is the adjustment factor, ΔP(B i |F) is the change in the probability of the behavior occurring, and N is the number of root nodes; The step of inferring the user's next operation according to the behavior prediction model and generating a pre-fetch instruction includes calculating the priority of the pre-fetched data block. The method of calculating the priority of the pre-fetched data block is as follows: , where P(D i ) is the data block D i The priority score, δ1, δ2 and δ3 are weight coefficients, T now is the current time, T last (D i ) is the last access time of the data block, T total is the total time unit, F access (D i ) is the access frequency of data blocks, F total is the total access frequency, S(D i ) is the size of the data block, S total is the total data block size; The calculation method of the cache replacement is: , where R(D i ) is the data block D i The replacement priority, λ, η and σ are weight factors, T recency (D i ) is the last access time of the data block, F frequency (D i ) is the access frequency, S(D i ) is the size of the data block.

2. The method for prefetching data with intent-awareness for a solid-state storage device according to claim 1, wherein: The step of collecting user behavior data includes: Capture and record user operations in different applications, including opening applications, editing files, saving and closing; Record access logs in the file system, including file creation, modification, deletion, and read and write operations; Monitor and record users' network activities, including traffic volume, accessed IP addresses, and access time.

3. The intent-aware data pre-fetching method for a solid-state storage device according to claim 1, wherein: The steps of classifying, filtering, and preprocessing the collected user behavior data, generating time series and behavior series, and mapping them into multi-dimensional feature vectors include: Convert the user behavior data into a unified format and filter out noise and outliers; Extracting features from the processed user behavior data to obtain feature data, wherein the feature data includes time features, frequency features, data type features, user interaction mode features, and network access features; The feature data are assembled into a multi-dimensional feature vector and vectorized according to specific requirements.

4. The method for pre-fetching data with intent-awareness for a solid-state storage device according to claim 1, wherein: The step of pre-fetching target data from different storage layers of the solid-state storage device according to the pre-fetch instruction and storing the target data in a designated cache area according to priority includes: Based on the priority list in the prefetch instruction, selectively reading a target data block from the solid-state storage device for data block location; Transmitting the selected target data block to the buffer area in segments as needed to prevent buffer overflow and data transmission blockage; The data storage structure in the cache area is dynamically adjusted according to the priority list in the prefetch instruction to ensure that high priority data can be quickly accessed.

5. The method for pre-fetching data with intent-awareness for a solid-state storage device according to claim 1, wherein: The method further includes user feedback correction, which performs feedback correction on the actual operation result of the user and adjusts the pre-fetching decision model to improve the pre-fetching accuracy. The user feedback correction step includes: Statistics and analysis of pre-fetched data hits and cache hit rates; Dynamically adjust user behavior characteristics and model parameters based on user operation feedback to optimize the user intention prediction model; Incremental learning and / or online learning algorithms are used to continuously optimize and enhance the pre-fetching decision model.

6. An intention-aware data pre-fetching device for a solid-state storage device, used to implement the intention-aware data pre-fetching method for a solid-state storage device according to any one of claims 1 to 5, characterized in that: The device comprises: The data collection module is used to collect application usage frequency, file access time, system call logs, hard disk read traces and network traffic information through a multi-source data collection mechanism; A multi-dimensional behavior analysis module processes and analyzes the data obtained from the data acquisition module, generates a time series and a behavior sequence, and maps the time series and the behavior sequence into a multi-dimensional feature vector; User behavior modeling module, based on multi-dimensional feature vectors, uses a comprehensive reasoning decision tree algorithm to build a user behavior prediction model; The pre-fetch decision generation module uses the user behavior prediction model to infer the user's next operation and generate pre-fetch decision information and priority list; A data pre-fetch and cache module, used to pre-fetch required data blocks from the solid-state storage device and store them in a cache area according to the pre-fetch decision information; The cache and data management module is used to dynamically manage the cache area by using elastic cache size adjustment and cache replacement.

7. A non-volatile readable storage medium for implementing intent-aware data prefetching, characterized in that: The storage medium stores computer-readable instructions, which, when executed by a computer, enable the computer to implement the intention-aware data pre-fetching method for a solid-state storage device as described in any one of claims 1 to 5.

Images