Generation method and correction method for computer memory access record of external equipment

Abstract

The invention discloses a generation method and a correction method for a record of external equipment accessing a computer memory. The correction method comprises the following steps: extracting feature vectors from each picture in a real data set and generation content needing to be corrected; searching a sample most similar to the generated content feature vector in a feature vector space of the real data set; calculating the distance between the correction basis and the generated content at each pixel position to form a distance vector; based on the distance vector, calculating the singularity of each pixel by using a predetermined weight so as to evaluate the difference degree between the singularity and the pixel at the corresponding position of the correction base; traversing the singularity vector according to a hyper-parameter given by a user, and selecting original pixels in the reserved generated content or corresponding pixels in the correction basis for replacement; and outputting the generated data as a final result when the generated data meets the requirements of the user after being subjected to iteration correction for multiple times. According to the method, the generation quality and precision of the high-performance peripheral memory access record of the computer are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of peripheral device access record generation and correction, and particularly relates to a method for generating and correcting a record of peripheral device access to computer memory. BACKGROUND

[0002] In the architecture of modern computer systems, high-performance peripherals such as graphics cards (GPU), solid-state drives (SSD), and network adapters, as data processing accelerators, require data transmission between the host to have extremely high bandwidth and low latency characteristics. This demand has driven the popularity of PCIe (Peripheral Component Interconnect Express) as the mainstream connection standard, as it not only provides data transmission rates far exceeding traditional bus standards, but also ensures the efficiency and parallelism of data transmission through its unique point-to-point topology and supported multiple data channels.

[0003] The introduction of PCIe enables high-performance peripherals to fully utilize their internal high-speed processing capabilities and large-capacity data throughput potential, meeting the extreme demands of data-intensive applications such as big data analysis, high-definition video processing, real-time game rendering, and other data-intensive applications for computing and storage resources. This seamless connection not only improves the overall response speed and efficiency of the system, but also promotes the deep integration between hardware and software, driving further development of computer technology.

[0004] However, to fully exploit the potential of these high-performance peripherals, it is particularly important to accurately acquire and manage the access information of PCIe devices. Access information covers detailed records of data exchange between peripherals and main memory, which plays an irreplaceable role in optimizing data transmission paths, reducing data latency, and improving system throughput. At the same time, through analysis of access information, system administrators can more scientifically allocate and manage memory resources, preventing performance bottlenecks and security issues caused by resource conflicts or overloads.

[0005] However, in actual operation, directly collecting access information from PCIe devices often faces many challenges. On the one hand, real-time data collection may interfere with the normal operation of the system, leading to performance degradation or even system crashes. On the other hand, in some cases, due to the scarcity or limitations of hardware samples, direct data collection becomes impractical. Therefore, synthetic access records have become a viable alternative, avoiding the complexity and noise interference in the actual data collection process. By simulating access records that meet the characteristics of PCIe devices and interconnection networks, it is possible to provide strong data support for performance optimization, resource management, and security analysis without interfering with the normal operation of the system.

[0006] However, it is worth noting that simple synthesis methods often cannot guarantee the authenticity and effectiveness of the generated data. In order to ensure that the synthesized access record can truly reflect the actual running situation and have practical application value, it is necessary to deeply study the internal mechanism and working principle of PCIe device and interconnection network, and ensure that the generated content strictly follows the relevant standards and specifications. Only in this way, can we provide a solid data foundation for the performance improvement, resource optimization and security protection of computer system. SUMMARY

[0007] Therefore, the present application provides a method for generating and correcting computer memory access records by external devices, to solve the problem of improving the accuracy and effectiveness of synthesized PCIe access records in the prior art.

[0008] The specific technical solutions of the present application are as follows:

[0009] A method for generating computer memory access records by external devices, comprising:

[0010] Step S1, collecting original data of access activities from external devices, and cleaning the collected data to remove noise, outliers and duplicate records, to ensure the accuracy and integrity of the data; the cleaned data is selected and constructed into a basic data set;

[0011] Step S2, according to the application requirements, selecting a suitable generative artificial intelligence model, using the constructed basic data set to train or fine-tune the selected model, to optimize the model parameters and improve the quality and precision of the generated results;

[0012] Step S3, according to the specific requirements of the target software and hardware environment, designing a set of prompt words, which accurately reflect the characteristics of the environment and the specific requirements of the access record; providing the designed prompt words as input to the trained generative artificial intelligence model, and the model generates preliminary access record data according to the prompt words; the generated original access data includes text and image forms;

[0013] Step S4, performing regularization processing on the generated original access data;

[0014] Step S5, the generated record of external device accessing computer memory is presented in the form of image, with uniform size, clear quality and accurate information expression, to be used in the development and debugging of high-performance computer system, providing intuitive and convenient data analysis means.

[0015] Further, the original data in step S1 includes read and write requests, address information and time stamp.

[0016] Further, the generative artificial intelligence model in step S2 is an adversarial generative network, a variational autoencoder, a diffusion model, a long short-term memory network, or a large language model.

[0017] Further, in step S4, for the original access data in the form of an image, an image processing algorithm is used to crop, adjust the image to a uniform size, and perform basic image denoising to improve image quality.

[0018] Further, in step S4, for the original access data in the form of text, a specific encoding and decoding method is used to extract key information from the text data, discard irrelevant data fields, reorder the entries in the record according to time sequence, and encode each entry as an RGB pixel point. These RGB pixel points are arranged in the order of the record to form an image.

[0019] A correction method for external device access to computer memory records, comprising:

[0020] Step a, receiving the user-provided generated content to be corrected, verifying whether it has completed regularization processing, and ensuring that the real data set used for correction is consistent in format with the generated content;

[0021] Step b, using a feature extractor based on an artificial intelligence model or a traditional mathematical method to extract feature vectors from each picture in the real data set and the generated content to be corrected, which sufficiently reflect the essential features of the data;

[0022] Step c, using an efficient search algorithm to find the most similar sample to the feature vector of the generated content in the feature vector space of the real data set, and using it as the basis for subsequent correction;

[0023] Step d, calculating the distance between the basis for correction and the generated content at each pixel position to form a distance vector; based on the distance vector, using a predetermined weight to calculate the singularity of each pixel to evaluate the difference between it and the corresponding pixel of the basis for correction; according to the user-given hyperparameters, traverse the singularity vector, and select to keep the original pixel in the generated content or replace it with the corresponding pixel in the basis for correction;

[0024] Step e, according to user requirements, the generated content is iteratively corrected multiple times; after each iteration, the quality of the generated data is re-evaluated to determine whether it meets the preset correction standard; if not, return to the feature extraction step for a new round of correction process;

[0025] Step f, when the generated data meets the user's requirements after multiple iterations of correction, it is output as the final result, and the output data reflects the external device's access to the computer's memory.

[0026] Furthermore, in step a, the regularization process includes standardizing the data format, adjusting the resolution, converting the color space, and removing noise.

[0027] Furthermore, in the matching process of step c, cosine similarity or Euclidean distance is used as a similarity metric.

[0028] Furthermore, in step d, the hyperparameters given by the user include the singularity threshold and the replacement ratio.

[0029] Furthermore, in step e, a user feedback mechanism is introduced to adjust hyperparameters or correction strategies based on users' intuitive feelings.

[0030] The beneficial effects of this invention are as follows:

[0031] 1. Efficient basic data preprocessing: This method first emphasizes the rigorous cleaning and careful selection of the collected basic data. This step not only removes irrelevant and noisy data, but also retains the key information that best reflects the characteristics of device memory access behavior, providing a high-quality basic dataset for subsequent model training.

[0032] 2. Intelligent Model Generation: Utilizing a preprocessed base dataset, this invention trains or fine-tunes a generative artificial intelligence model. This model is capable of learning and simulating complex patterns of device memory access behavior based on a large amount of real data, thereby generating preliminary but highly accurate memory access records. This process demonstrates the innovative application of artificial intelligence technology in the field of data generation.

[0033] 3. Flexible Data Regularization: To further improve the quality of the generated data, this invention introduces a data regularization step. For raw image data, cropping and noise reduction are performed; for text data, a unique encoding and decoding method is used to map it into RGB pixels and arrange them into an image in chronological order. This step ensures the uniformity of the generated data format and its visual readability, laying the foundation for subsequent correction work.

[0034] 4. User-Customized Correction Strategy: A major highlight of this method is its user-defined hyperparameter-based correction approach. Users can adjust the correction parameters according to actual needs to guide further correction of the generated data. Through techniques such as feature vector matching, pixel distance calculation, and singularity analysis, this invention can accurately identify the differences between the generated data and the real data, and accordingly perform targeted pixel replacement or retention, thereby significantly improving the quality of the generated data.

[0035] 5. Adaptive Generation and Correction Mechanism: This invention also designs an adaptive memory access record generation and correction method, specifically for devices represented by high-performance PCIe peripherals. This method can flexibly adjust the generation and correction strategies according to different application scenarios and constraints to adapt to different hardware and software environment requirements. This adaptability not only improves the efficiency and accuracy of data generation but also enhances the broad applicability of this invention in practical applications.

[0036] 6. Broad Application Prospects: This invention is not only applicable to the generation and correction of memory access records for various high-performance peripherals, but can also be widely used in the development and debugging of high-performance computer systems. Its efficient and accurate methods for generating and correcting memory access record data provide strong support for technological advancements and industrial upgrading in related fields. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the text-based visual encoding and decoding process in the method of this invention;

[0039] Figure 2 This is a schematic diagram of the data regularization process in the generation method of the present invention;

[0040] Figure 3 This is a flowchart illustrating the data correction method based on user-given hyperparameters in the correction method of this invention. Detailed Implementation

[0041] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0042] This invention proposes a method for generating and correcting records of external device access to computer memory. This method is based on generative artificial intelligence. The invention is meticulously designed to significantly improve the quality and accuracy of memory access records generated from high-performance computer peripherals (such as graphics cards, solid-state drives, and network adapters). The core of this invention lies in the rigorous cleaning and fine selection of raw basic data collected from various high-performance peripherals, eliminating invalid, erroneous, or redundant data to construct a high-quality basic dataset. This dataset is then used as input for training or fine-tuning a generative artificial intelligence model that can learn and simulate the memory access behavior patterns of real devices, thereby generating preliminary memory access records.

[0043] After generating initial memory access records, this invention employs data regularization techniques. Through a series of mathematical transformations and filtering processes, the generated data becomes smoother and more stable, reducing noise interference. Subsequently, based on user-defined hyperparameters (such as error tolerance and optimization objectives), this invention implements a refined correction method to further optimize and adjust the regularized data, ensuring it accurately reflects the memory access behavior characteristics of the actual device. This process not only improves data accuracy but also enhances data reliability and usability.

[0044] Another major highlight of this invention lies in its innovative application in distributed virtual machine environments. In this context, the invention designs a novel adaptive consistency protocol that dynamically adjusts memory consistency strategies based on the system's real-time state and load, thereby significantly improving the performance and efficiency of distributed shared memory. Simultaneously, by utilizing a simulated Global Memory Management Unit (gMMU) in the guest machine, the invention successfully incorporates DMA access from device-pass-through I / O devices into the management scope of distributed shared memory, achieving fine-grained control and optimization of DMA access.

[0045] In summary, this invention not only provides efficient and accurate methods for generating and correcting memory access record data for the development and debugging of high-performance peripherals, but also further improves the overall performance and reliability of the system in a distributed virtual machine environment by introducing innovative technologies such as adaptive consistency protocols and gMMU-managed DMA access. This method is particularly suitable for devices represented by PCIe high-performance peripherals, and can flexibly adjust the generation and correction strategies under different application scenarios and constraints to meet the diverse needs of the development and debugging of high-performance computer peripheral related software and hardware systems.

[0046] This invention proposes a method for generating records of external device access to computer memory, aiming to simulate and predict the memory access behavior of peripheral devices under different hardware and software environments through intelligent means. The entire process includes multiple steps such as basic data collection and cleaning, generative AI model training and fine-tuning, prompt word design, and data generation and regularization.

[0047] First, basic peripheral memory access data is collected using specialized data acquisition tools in a specified hardware and software environment. This raw data undergoes rigorous cleaning to remove noise and outliers, ensuring its accuracy and reliability. Then, the cleaned dataset is used to train or fine-tune generative artificial intelligence models. These models include, but are not limited to, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models, Long Short-Term Memory Networks (LSTMs), and large language models, to capture the inherent patterns and features within the data.

[0048] After model training is complete, corresponding prompts are designed based on the characteristics of the target hardware and software environment. These prompts serve as input to the generative AI model, guiding it to generate raw memory access data that meets specific requirements. The generated raw memory access data may exist in the form of text or images. For raw image data, cropping and denoising will be performed to ensure that the image quality meets the needs of subsequent analysis or applications. For raw text data, such as... Figure 1 As shown, a specialized encoding and decoding method is used to encode the text data into RGB pixels and arrange them in chronological order to generate an image, thereby realizing the visualization of the text data.

[0049] To further improve the standardization and usability of the data, the generated raw data also needs to be regularized. The regularization process employs different strategies depending on the different forms of the data, such as... Figure 2 As shown: For image data, cropping and resizing to a uniform size, followed by basic noise reduction; for images converted from text data, mapping and arranging are performed according to a specific encoding and decoding process, ultimately forming image data with a uniform format and clear content. After this series of processing steps, the final product is regularized content, which can be directly used for subsequent data analysis, model training, or application demonstrations.

[0050] This invention proposes a method for correcting records of external devices accessing computer memory. This method uses user-defined hyperparameters to finely control the correction process of the generated data, ensuring that the generated content more closely resembles the characteristics of real data. For example... Figure 3 As shown, the specific steps are as follows:

[0051] First, this method requires that the generated content has already undergone regularization to ensure data format consistency and quality. Simultaneously, the real dataset used for correction also needs to maintain a consistent format with the generated content for subsequent feature extraction and matching operations.

[0052] Next, feature extractors based on artificial intelligence models (such as deep learning networks) or traditional mathematical methods (such as feature descriptors) are used to extract embedding vectors for all images and the content to be corrected in the real dataset. These embedding vectors capture the key features of the data, providing a foundation for subsequent feature matching.

[0053] During the feature matching stage, the distance between the generated content and the embedding vectors of each sample in the real dataset (such as Euclidean distance, cosine similarity, etc.) is calculated to find the closest real data as the basis for correction. The purpose of this step is to find a reliable reference point for the generated content in order to perform subsequent pixel-level correction.

[0054] Subsequently, the distance between each corresponding pixel in the modified base and the generated content is calculated according to preset weights, forming a distance vector. Then, this distance vector is converted into a pixel singularity vector using another weight. Singularity reflects the degree of "abnormality" of each pixel in the generated content compared to the modified base.

[0055] During the pixel correction stage, the singularity vector is traversed, and based on user-provided hyperparameters (such as singularity threshold, replacement intensity, etc.), the system selects to retain pixels at the corresponding positions in the generated content or replace them with pixels at the same positions in the correction base. This step allows users to flexibly control the degree and effect of correction according to their actual needs.

[0056] Finally, if the generated data still does not meet the user's needs after one correction, the above correction process (starting from feature extraction) can be repeated until the generated data achieves a satisfactory correction effect. This process allows users to continuously optimize the quality of the generated data by adjusting hyperparameters.

[0057] In summary, this correction method guides the correction process of generated data through user-given hyperparameters, and combines steps such as feature extraction, matching, and pixel-level correction to achieve effective correction and optimization of peripheral memory access records generated by generative artificial intelligence models.

[0058] A detailed deployment example of this invention is as follows: A personal computer is selected as the deployment platform. This computer is equipped with a stable and feature-rich operating system, Ubuntu Desktop 22.04.5LTS 64-bit, and is connected to a target PCIe high-performance network card, ensuring efficient and real-time data transmission. Furthermore, the computer is also equipped with a Realtek 8169 Gigabit network interface card, providing a solid foundation for network communication.

[0059] To construct an accurate generative model, this invention utilizes specialized tools to conduct detailed data collection on the target high-performance peripherals within a specified hardware and software environment. The collected raw data undergoes rigorous cleaning and preprocessing, ultimately yielding a high-quality, realistic dataset. This dataset serves as the foundation for model training and is crucial for improving the accuracy and realism of the generated data.

[0060] In terms of model selection, this invention employs a variety of advanced spindle generative models, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Diffusion Models, Long Short-Term Memory (LSTMs), and large language models. These models each have their own characteristics and can be flexibly selected and combined according to the needs of different tasks. Simultaneously, a feature extractor, including pre-trained and specially trained AI models, is provided to extract key features useful for the generation task from complex data. Notably, the system architecture of this invention has good compatibility, supporting the integration and expansion of other spindle generative models and feature extractor configurations.

[0061] In practical applications, the system can easily adapt to different combinations of computing devices and high-performance peripherals. Whether it's connecting a 10 Gigabit network interface card to improve network transmission speed, connecting a high-performance graphics processor to accelerate graphics processing tasks, or using a solid-state drive to provide a high-speed data storage solution, the system can be adjusted and optimized to meet the needs of different scenarios.

[0062] For specific generation tasks, users can achieve fine-grained control over the generated data by modifying the model's hyperparameter configuration and specifying a real dataset. This interpretable and guideable generated data correction mechanism allows users to flexibly adjust and optimize the generation process according to their needs and goals. Simultaneously, by continuously adjusting and optimizing the structure and parameter settings of the backbone generative AI model, the system can achieve varying degrees of raw data generation capability for different tasks. This flexibility enables the system to be widely applied in various complex data generation and correction scenarios.

[0063] In summary, the system of this invention integrates multiple advanced spinal column generative models and feature extractors, supports flexible device configuration and generation task adjustments, and provides an interpretable and guided generation data correction mechanism, thereby achieving the generation and interpretable correction of device memory access records for specific high-performance peripherals. This innovative solution brings entirely new ideas and methods to the field of data processing and generation.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating an external device access computer memory record, characterized by, The method comprises the following steps: Step S1, collecting raw data of memory access activities from external devices, cleaning the collected data to remove noise, outliers and duplicate records, and ensuring the accuracy and integrity of the data; the cleaned data is selected and constructed into a basic data set; Step S2, according to the application requirements, select a suitable generative artificial intelligence model, use the constructed basic data set to train or fine-tune the selected model, optimize the model parameters, and improve the quality and accuracy of the generated results; Step S3, according to the specific requirements of the target software and hardware environment, design a set of prompt words that accurately reflect the characteristics of the environment and the specific requirements of the memory record; provide the designed prompt words as input to the trained generative artificial intelligence model, and the model generates preliminary memory record data according to the prompt words; the generated raw memory data includes text and image forms; Step S4, regularize the generated raw memory data; Step S5, the generated record of external device accessing computer memory is presented in image form, with uniform size, clear image quality and accurate information expression, for the development and debugging of high-performance computer systems, providing intuitive and convenient data analysis means.

2. The method of claim 1, wherein the external device accesses the computer memory record. The raw data in step S1 includes read and write requests, address information, and timestamps.

3. The method of claim 1, wherein the external device accesses the computer memory record. The generative artificial intelligence model in step S2 is a generative adversarial network, a variational autoencoder, a diffusion model, a long short-term memory network, or a large language model.

4. The method of claim 1, wherein the method further comprises: generating a record of the access of the computer memory by the external device. In step S4, for image-form raw memory data, use image processing algorithms to crop and adjust the image to a uniform size, and perform basic image denoising to improve image quality.

5. The method of claim 1, wherein the method further comprises: generating a record of the access of the memory of the computer by the external device. In step S4, for text-form raw memory data, use a specific encoding and decoding method to extract key information from the text data, discard irrelevant data fields, reorder the entries in the record according to time, and encode each entry as an RGB pixel point. Arrange these RGB pixel points in the order of the record to form an image.

6. A correction method of an external device accessing a computer memory record, applied to the generation method of an external device accessing a computer memory record according to claim 1, characterized in that, The correction method comprises: Step a, receiving user-provided generated content to be corrected, verifying whether it has completed regularization processing, and ensuring that the real data set used for correction is consistent in format with the generated content; Step b, using a feature extractor based on an artificial intelligence model or a traditional mathematical method to extract feature vectors that sufficiently reflect the essential features of the data from each image in the real data set and the generated content to be corrected; Step c, using an efficient search algorithm to find the most similar sample to the feature vector of the generated content in the feature vector space of the real data set, and using it as the basis for subsequent correction; Step d, calculate the distance between the correction basis and the generated content at each pixel position to form a distance vector; based on the distance vector, calculate the singularity of each pixel using a predetermined weight to evaluate the difference between the corresponding pixels of the correction basis; according to the user-provided hyperparameters, traverse the singularity vector, and select to keep the original pixels in the generated content or replace them with the corresponding pixels in the correction basis. Step e, according to the user's demand, the generated content is iteratively corrected for multiple times; after each iteration, the quality of the generated data is re-evaluated to determine whether the preset correction standard is met; if not, return to the feature extraction step for a new round of correction process; Step f, when the generated data meets the user's demand after multiple iterations, it is output as the final result, and the output data reflects the access record of the external device to the computer memory.

7. The method of claim 6, wherein the external device accesses the computer memory record, and In step a, the regularization process includes data format unification, resolution adjustment, color space conversion and noise removal.

8. The method of claim 6, wherein the external device accesses the computer memory record, and In the matching process of step c, cosine similarity or Euclidean distance is used as the similarity measure.

9. The method of claim 6, wherein the external device accesses the computer memory record, and In step d, the user-given hyperparameters include the singularity threshold and the replacement ratio.

10. The method of claim 6, wherein the external device accesses the computer memory record, and In step e, a user feedback mechanism is introduced to adjust the hyperparameters or correction strategy according to the user's intuitive feeling.

Images