Parameter recommendation method and device

Abstract

Disclosed are a parameter recommendation method and device, relating to the computer field, for recommending optimal parameters under different performance conditions in various application scenarios, thereby improving the effectiveness and efficiency of parameter recommendation. The method includes: obtaining a tuning task; collecting and analyzing performance data of the object to be tuned to obtain a performance report for the object to be tuned; and, based on the performance report, searching a parameter knowledge base for the object to be tuned to determine a parameter set.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a parameter recommendation method and device. Background Art

[0002] Configurable parameters are predefined variables or options within an operating system (OS) or application that users or administrators can adjust based on specific needs. These parameters directly impact system performance, stability, security, and functionality. Adjusting configurable parameters can control system behavior, resource allocation, or functional characteristics. The process of adjusting configurable parameters to optimize performance, efficiency, or achieve specific goals is called parameter tuning.

[0003] As hardware and software scale, the number of adjustable parameters in modern operating systems and the applications running on them has skyrocketed, far exceeding the capabilities of manual tuning. Traditional manual tuning methods that rely on expert experience are no longer able to cope with complex and changing scenario requirements. A more intelligent parameter tuning process is currently being pursued. This intelligent parameter tuning process introduces intelligent models to process massive amounts of data and extract deep features to automatically select and recommend tuning parameters, optimize the search space, or generate high-quality parameter recommendations, improving the efficiency and effectiveness of parameter tuning.

[0004] However, many existing intelligent parameter optimization solutions still suffer from low parameter recommendation efficiency and accuracy. In addition, the versatility of parameter recommendation application scenarios is also a goal pursued by the industry. Summary of the Invention

[0005] This application provides a parameter recommendation method and device, which can recommend optimal parameters under different performance states in various application scenarios to improve the effect and efficiency of parameter recommendation.

[0006] In a first aspect, a parameter recommendation method is provided. The method may include: obtaining a tuning task including a target object and an optimization objective; collecting and analyzing performance data of the target object to obtain a performance report describing the current performance of the target object; and, based on the performance report, searching a parameter knowledge base for the target object to determine a parameter set, the parameter set including one or more configurable parameters that affect the optimization objective. The parameter knowledge base includes knowledge information for adjusting the configurable parameters of the target object.

[0007] In the solution provided by this application, the performance data of the object to be adjusted is collected and analyzed to generate a performance report reflecting the performance problems of the object to be adjusted. The parameter knowledge base is then searched with reference to the performance report to determine the parameter set for parameter recommendation, replacing the current solution of parameter recommendation based on the full input model. Since the performance report of the object to be adjusted is combined in the process of determining the parameter set, the configurable parameters in the determined parameter set are strongly correlated with the current performance of the object to be adjusted, and the optimal tuning parameters under different performance states can be obtained, so that the quality of parameter recommendation is high and the effect is good.

[0008] In one possible implementation, the performance report includes bottleneck information and / or workload information. The bottleneck information describes the performance bottleneck of the target object, while the workload information describes the workload of the target object. The performance bottleneck and / or workload of the target object are considered during the search, ensuring that the recommended parameters retrieved are appropriate for the current bottleneck and / or workload status and match the current state of the target object, thereby ensuring the accuracy of the recommended parameters. This ensures that subsequent use of the recommended parameters (e.g., parameter tuning) is more efficient and effective.

[0009] In another possible implementation, the performance report includes workload information, and the tuning task also includes load description information. The load description information indicates the workload of the object being tuned. Accordingly, collecting and analyzing the performance data of the object being tuned to obtain the performance report for the object being tuned includes using the load description information as a prompt to analyze the performance data and obtain workload information in the performance report. The workload information is obtained based on user instructions, improving the accuracy of the obtained workload information and thereby ensuring the accuracy of parameter recommendations.

[0010] Another possible implementation method is to search the parameter knowledge base based on the performance report, including: searching the parameter knowledge base based on the performance report using semantic and / or keyword search methods. By using a hybrid search method, the advantages of each search method are utilized to improve the accuracy of the search results.

[0011] Another possible implementation method of the present invention may further include obtaining a parameter knowledge base, which is a structured parameter library extracted from heterogeneous knowledge sources based on a hierarchical prompt chain. By decomposing complex extraction tasks into multiple levels of prompt chains and combining them with few-shot learning techniques to process parameter knowledge from various fields, the proposed solution is generalizable and the acquired knowledge information can improve the effectiveness and efficiency of recommendations.

[0012] Another possible implementation involves obtaining a parameter knowledge base. This involves obtaining parameter tuning information for the target object from multiple sources; extracting parameter knowledge from each source using a large language model (LLM) based on a hierarchical prompt chain; and fusing this parameter knowledge to form a parameter knowledge base. This structured parameter base is extracted from heterogeneous knowledge sources using a hierarchical prompt chain. This decomposes the complex extraction task into multiple levels of prompt chains. Combined with few-shot learning techniques, this system gradually guides the large model through knowledge organization, extraction, and integration, improving the large model's task-solving capabilities and the generalizability of knowledge extraction. This ensures that inputting the knowledge information from the large model during parameter recommendation improves both effectiveness and efficiency. Furthermore, this approach to constructing the knowledge parameter base offers a high degree of automation and can handle parameter knowledge from a variety of domains, ensuring the generalizability of the solution.

[0013] Another possible implementation involves fusing parameter knowledge from various sources to create a parameter knowledge base. Specifically, this involves determining the confidence level of each field in the parameter knowledge from each source; this confidence level describes the trustworthiness of the field; constructing fusion prompts based on the confidence level; and selecting the field with the highest confidence level based on the fusion prompts to fuse the parameter knowledge from various sources to create the parameter knowledge base. Using a confidence-aware fusion mechanism helps large models prioritize fields within each source, improve the accuracy of parameter knowledge base results, and resolve multi-source data conflicts, as well as the issues of insufficient field integrity and low overall accuracy faced by existing solutions.

[0014] Another possible implementation method of the parameter recommendation method provided herein may further include adjusting the values ​​of configurable parameters in a parameter set until a termination condition is met, where the termination condition includes the performance of the object to be tuned meeting the optimization objective. The recommended parameter results are used for parameter tuning. Due to their high accuracy and efficiency, the recommended parameter results can effectively improve the effectiveness and efficiency of parameter tuning.

[0015] In another possible implementation, the termination condition further includes that the number of iterations is greater than or equal to a threshold.

[0016] Secondly, a method for constructing a parameter knowledge base is provided, which specifically includes: obtaining parameter tuning information for an object to be tuned from multiple sources; extracting parameter knowledge from each source from the parameter tuning information based on a hierarchical prompt chain using LLM; and fusing the parameter knowledge from each source to obtain a parameter knowledge base.

[0017] Based on a hierarchical prompt chain, a structured parameter library is extracted from heterogeneous knowledge from multiple sources. This breaks down the complex extraction task into a multi-level prompt chain. Incorporating few-shot learning techniques, this gradually guides the large model through knowledge organization, extraction, and integration, improving both its task-solving capabilities and the generalizability of knowledge extraction. This ensures that inputting the knowledge information of the large model during parameter recommendation improves both effectiveness and efficiency. Furthermore, this method for constructing the knowledge parameter library is highly automated and can handle parameter knowledge from a variety of fields, ensuring the generalizability of the solution.

[0018] In one possible implementation, the method for constructing a parameter knowledge base provided in this application may further include: obtaining a tuning task including an object to be tuned and an optimization target; collecting and analyzing performance data of the object to be tuned to obtain a performance report describing the current performance of the object to be tuned. The parameter knowledge base for the object to be tuned is searched to determine a parameter set, where the parameter set includes one or more configurable parameters that affect the optimization target. The constructed parameter knowledge base is applied to parameter recommendation scenarios, making the application scenarios of the parameter recommendation solution more extensive.

[0019] In another possible implementation, the method for constructing a parameter knowledge base provided by the present application may further include: collecting and analyzing the performance data of the object to be adjusted, and obtaining a performance report for describing the current performance of the object to be adjusted. Accordingly, the parameter knowledge base for the object to be adjusted is searched to determine a parameter set, which is specifically implemented as follows: based on the performance report, the parameter knowledge base for the object to be adjusted is searched to determine a parameter set. By collecting and analyzing the performance data of the object to be adjusted, a performance report reflecting the performance problems of the object to be adjusted is generated. The parameter knowledge base is then searched with reference to the performance report to determine a parameter set for parameter recommendation, replacing the current solution of recommending parameters using a full input model. Since the performance report of the object to be adjusted is combined in the process of determining the parameter set, the configurable parameters in the determined parameter set are strongly correlated with the current performance of the object to be adjusted, and the optimal tuning parameters under different performance states can be obtained, so that the quality of the parameter recommendation is high and the effect is good.

[0020] The process of searching the parameter knowledge base for the object to be adjusted based on the performance report and determining the parameter set can refer to the relevant content of the first aspect above and will not be repeated here.

[0021] Another possible implementation method of the parameter knowledge base construction method provided herein may further include adjusting the values ​​of configurable parameters in a parameter set until a termination condition is met, where the termination condition includes the performance of the object to be tuned meeting the optimization objective. The recommended parameter results are used for parameter tuning. Due to their high accuracy and efficiency, the recommended parameter results can effectively improve the effectiveness and efficiency of parameter tuning.

[0022] In a third aspect, a parameter recommendation device is provided, which may include: an acquisition unit and a parameter recommendation unit.

[0023] The acquisition unit is used to acquire the tuning task, which includes the object to be tuned and the optimization target.

[0024] The parameter recommendation unit is used to collect and analyze the performance data of the target object to generate a performance report describing the current performance of the target object. Based on the performance report, the parameter knowledge base for the target object is searched to determine a parameter set. The parameter set includes one or more configurable parameters that affect the optimization objective. The parameter knowledge base contains knowledge information used to adjust the configurable parameters of the target object.

[0025] The parameter recommendation device provided in the third aspect is used to execute the method described in the first aspect or any possible implementation of the first aspect. Its specific implementation and beneficial effects can refer to the contents described in the first aspect or any possible implementation of the first aspect, and will not be repeated here.

[0026] In a fourth aspect, a device for constructing a parameter knowledge base is provided, which may include: an acquisition unit, an extraction unit, and a fusion unit.

[0027] The acquisition unit is used to acquire parameter tuning information for the object to be tuned from multiple sources.

[0028] The extraction unit is used to extract parameter knowledge from various sources from parameter tuning information based on a hierarchical hint chain using LLM.

[0029] The fusion unit is used to fuse the parameter knowledge from various sources to obtain a parameter knowledge base.

[0030] The device for constructing a parameter knowledge base provided in the fourth aspect is used to execute the method described in the second aspect or any possible implementation of the second aspect. Its specific implementation and beneficial effects can refer to the contents described in the second aspect or any possible implementation of the second aspect, and will not be repeated here.

[0031] In a fifth aspect, a computing device is provided, comprising a processor and a memory. The processor is configured to execute instructions stored in the memory, so that the computing device performs the steps of the method of the first aspect, the second aspect, or any possible implementation thereof.

[0032] In a sixth aspect, a computer program product comprising instructions is provided, which, when executed by a computing device, causes the computing device to perform the operating steps of the method described in the first aspect or the second aspect or any possible implementation manner.

[0033] In the seventh aspect, a computer-readable storage medium is provided, comprising: computer program instructions; when the computer program instructions are executed by a computing device, the computing device performs the operating steps of the method described in the first aspect or the second aspect or any possible implementation.

[0034] The technical effects brought about by any design method in the third to seventh aspects can refer to the technical effects brought about by different design methods in the first or second aspects, and will not be repeated here.

[0035] Based on the implementation methods provided in the above aspects, this application can also be further combined to provide more implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of the architecture of a computer system provided in an embodiment of the present application;

[0037] Figure 2 A schematic diagram of the architecture of another computer system provided in an embodiment of the present application;

[0038] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application;

[0039] Figure 4 A flow chart of a parameter recommendation method provided in an embodiment of the present application;

[0040] Figure 5 A flow chart of another parameter recommendation method provided in an embodiment of the present application;

[0041] Figure 6 A schematic diagram of the principle of a parameter recommendation process provided in an embodiment of the present application;

[0042] Figure 7 A flow chart of a method for constructing a parameter knowledge base provided in an embodiment of the present application;

[0043] Figure 8 A schematic diagram of the principle of a method for constructing a parameter knowledge base provided in an embodiment of the present application;

[0044] Figure 9 A flow chart of another parameter recommendation method provided in an embodiment of the present application;

[0045] Figure 10 A schematic diagram of a specific application scenario of a parameter recommendation method provided in an embodiment of the present application;

[0046] Figure 11 A schematic diagram of a software implementation form of a server provided in an embodiment of the present application;

[0047] Figure 12 A schematic diagram of a multi-source heterogeneous data collection and preprocessing process provided in an embodiment of the present application;

[0048] Figure 13 A schematic diagram of a process for structured parameter extraction and knowledge supplementation provided in an embodiment of the present application;

[0049] Figure 14 A schematic diagram of a process of multi-source knowledge fusion with confidence priority provided in an embodiment of the present application;

[0050] Figure 15 A schematic diagram of the process of parameter knowledge vectorization and index construction provided in an embodiment of the present application;

[0051] Figure 16 A schematic diagram of a load-aware search enhancement process provided by an embodiment of the present application;

[0052] Figure 17 A schematic diagram of a process for prompt generation and parameter recommendation provided in an embodiment of the present application;

[0053] Figure 18 A schematic diagram of the structure of a parameter recommendation device provided in an embodiment of the present application;

[0054] Figure 19 A schematic diagram of the structure of another parameter recommendation device provided in an embodiment of the present application;

[0055] Figure 20 A schematic diagram of the structure of an apparatus for constructing a parameter knowledge base provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In the description of this application, unless otherwise specified, " / " indicates that the objects associated before and after are in an "or" relationship, for example, A / B can represent A or B; "and / or" in this application is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural.

[0057] In this application, unless otherwise specified, "plurality" refers to two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0058] In addition, to facilitate the clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0059] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner to facilitate understanding.

[0060] It is understood that the "embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the various embodiments in the entire specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in multiple embodiments in any suitable manner. It is understood that in the various embodiments of the present application, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0061] It is understood that some optional features in the embodiments of the present application may, in certain scenarios, be implemented independently of other features, such as the solution on which they are currently based, to solve corresponding technical problems and achieve corresponding effects. They may also be combined with other features in certain scenarios as needed. Accordingly, the devices provided in the embodiments of the present application may also implement these features or functions accordingly, which will not be described in detail here.

[0062] In this application, unless otherwise specified, the same or similar parts between the various embodiments can be referenced to each other. In the various embodiments of this application, unless otherwise specified and there is no logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships. The following description of the embodiments of this application does not constitute a limitation on the scope of protection of this application.

[0063] It should be noted that the information (including but not limited to device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the subject or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0064] To facilitate understanding, the main terms involved in this application are first explained.

[0065] Configurable parameters are predefined variables or options within an operating system (OS) or application that allow users or administrators to adjust them based on specific needs. Configurable parameters control system behavior, resource allocation, or functional characteristics. These parameters directly impact system performance, stability, security, and functionality.

[0066] Parameter tuning: Parameter tuning is the process of adjusting the configurable parameters of an operating system (OS) or applications (such as databases) running on it to optimize performance, efficiency, or achieve specific goals. It involves adjusting multiple configurable parameters that affect system behavior to find the optimal configuration, thereby improving key metrics such as system performance, responsiveness, and resource utilization.

[0067] Parameter recommendation: Parameter recommendation is a key step in the parameter tuning process. It involves filtering out a subset of key parameters that significantly impact target performance metrics (such as latency, throughput, and resource utilization) from a large number of configurable parameters based on historical tuning experience, real-time load characteristics, or predefined rules. Its core goal is to narrow the tuning scope and reduce the dimensionality of the search space, thereby improving subsequent tuning efficiency and effectiveness.

[0068] Parameter knowledge: This refers to knowledge that provides a detailed description of the nature, function, impact, interrelationships, and tuning strategies of configurable parameters. For example, parameter knowledge may include parameter types and definitions, parameter functions and impacts, parameter value ranges and types, interactions between parameters, tuning objectives, knowledge of tuning strategies and methods, rules of thumb and heuristics, and other related information. The content of parameter knowledge can be customized based on actual needs and is not limited by this application.

[0069] AI big models (abbreviated as "big models") refer to a class of AI models with a large number of parameters, constructed using artificial neural networks. They are typically pre-trained on massive amounts of data through self-supervised or semi-supervised learning, and their performance and capabilities are further optimized through methods such as instruction fine-tuning and human alignment. Big models are characterized by a large number of parameters, extensive training data, and extensive computing resources. They are capable of solving general tasks, following human instructions, and performing complex reasoning. Major categories of AI big models include big language models, big vision models, big multimodal models, and big basic science models. Currently, big models have been widely applied in multiple fields, driving the intelligent development of various industries.

[0070] Large Language Models (LLMs) are an AI technology based on deep learning. They are trained using large datasets, enabling them to generate natural language text or understand the meaning of text. Through a layered neural network structure, these models learn and simulate the complex patterns of human language, achieving near-human-level text generation capabilities.

[0071] Next, the current status of parameter tuning in the industry is described.

[0072] As the core of computing systems, the operating system's configurable parameters directly impact performance and stability in high-concurrency, high-load scenarios. Traditional parameter tuning typically relies on expert experience and manual tuning. However, with the expansion of hardware and software scale, the number of configurable parameters in modern operating systems and the applications running on them has skyrocketed, far exceeding the capabilities of manual tuning. Traditional manual tuning methods that rely on expert experience are no longer able to meet the complex and ever-changing requirements of scenarios.

[0073] To achieve more intelligent parameter tuning, the industry generally adopts the following approach: Automated tools that incorporate machine learning algorithms (such as Bayesian optimization and reinforcement learning) optimize parameter configurations through data-driven trial-and-error iteration. However, while this approach offers the advantage of automation, it still requires repeated parameter adjustments and performance testing within the environment, resulting in high computational resources and time costs. Furthermore, the algorithm lacks domain knowledge guidance and has weak generalization capabilities.

[0074] Against this backdrop, AI big models (hereinafter referred to as big models, such as large language models (LLMs)) offer a new approach to parameter tuning, leveraging their powerful knowledge understanding, learning, reasoning, and text generation capabilities. Specifically, they leverage the big model's capabilities (knowledge understanding, learning, reasoning, and text generation) to generate tuning prompts, which are then combined with machine learning algorithms for parameter tuning. This approach, enabled by the introduction of big models, can process massive amounts of parameter knowledge data and extract deep features from it. This allows for automated selection and recommendation of tuning parameters, optimizing the search space or generating high-quality parameter recommendations. This significantly reduces the inefficient trial-and-error process of existing methods, significantly improving tuning efficiency and effectiveness.

[0075] However, the training data for general-purpose large models is often extensive and unstructured, lacking a domain-specific parameter knowledge base and support for appropriate prompts. This results in low interpretability and accuracy in the parameter recommendation results obtained using large models. However, the industry currently lacks open-source and systematic parameter knowledge bases for various application ecosystems, and existing knowledge base construction methods are difficult to apply in multiple scenarios, making the initial knowledge collection and organization work for parameter tuning methods based on large models complex and cumbersome. Furthermore, this method uses a full knowledge input method, which causes the large model's attention mechanism to be dispersed to a large amount of low-relevance content, making the quality of generated recommendations suboptimal.

[0076] To solve these problems, the industry has designed the following solutions:

[0077] One solution is to process parameter knowledge through a large model to construct a structured knowledge view; adjust the parameter range through search space optimization technology to reduce the search space; and implement dynamic parameter tuning using a Bayesian optimization framework from coarse-grained to fine-grained. This solution can automatically extract domain knowledge through the power of a large model, reducing manual intervention in the construction of a parameter knowledge base; and use the constructed knowledge base as tuning suggestions during the parameter selection process, optimizing the parameter value space, accelerating the convergence speed of subsequent Bayesian optimization, and improving tuning efficiency. However, this solution uses a full knowledge input method, using all parameter knowledge as input to prompt the large model to recommend parameters, causing the model's attention mechanism to be distracted to a large amount of low-relevance content, thereby affecting the quality of the generated suggestions, resulting in low-quality parameter recommendations, and thus affecting the efficiency and effectiveness of subsequent tuning.

[0078] Another approach uses prompts designed to use the large model as an experienced database administrator to perform parameter tuning tasks. A solution, such as parameter recommendation based on the large model, is proposed as an alternative to traditional tuning methods. This approach uses the large model to analyze workload characteristics and identify key parameters with the greatest impact on performance. This narrows the search space, reduces unnecessary parameter adjustments, and improves tuning efficiency. Based on the parameter selection results (identified key parameters), the large model prompting project is used to initialize the model, providing an initial parameter configuration model and laying the foundation for subsequent iterative optimization. Finally, starting from this initial configuration, the large model is used to iteratively optimize based on database feedback (such as performance metrics), continuously adjusting the parameter configuration and performing parameter tuning to improve database performance. However, this approach uses unstructured parameter information as prompts for large model parameter selection. It is difficult for the large model to extract parameter value ranges and tuning suggestions from this unstructured information, resulting in suboptimal parameter recommendations, which affects the efficiency and effectiveness of subsequent tuning. Furthermore, this approach is designed only for database scenarios, with limitations in its input prompts for the large model. It lacks versatility in other application scenarios, limiting its applicability to a limited number of scenarios.

[0079] This shows that the efficiency and effectiveness of parameter tuning solutions currently available in the industry have not yet reached optimal levels. This is mainly due to the room for improvement in the efficiency and effectiveness of parameter recommendations, as well as the limited application scenarios of existing solutions.

[0080] Based on this, the present application provides a parameter recommendation method, which uses a parameter knowledge base for the object to be adjusted, combined with a performance report used to indicate the current performance status of the object to be adjusted, to search the parameter knowledge base to determine a parameter set of key parameters that affect performance, and perform parameter recommendations, replacing the current solution of parameter recommendation using a full input model. Since the performance report of the object to be adjusted is combined in the process of determining the parameter set, the configurable parameters in the determined parameter set are strongly correlated with the current performance of the object to be adjusted, and parameters under different performance states can be obtained, making the parameter recommendation high-quality and effective.

[0081] Furthermore, when this parameter recommendation solution is applied to the parameter tuning scenario, it can greatly improve the effect and efficiency of parameter tuning. Furthermore, in the present application solution, the performance report of the object to be tuned and the parameter knowledge base for the object to be tuned are combined when making parameter recommendations, thereby achieving retrieval enhancement, making up for the defect of insufficient utilization of domain knowledge by large models, and also enhancing the interpretability of parameter recommendations, thereby achieving a strong correlation between parameter recommendations and performance. Therefore, as long as the knowledge elaboration bases in different fields are configured, parameter tuning in various application scenarios can be achieved, avoiding the disadvantage of limited application scenarios.

[0082] Furthermore, current parameter recommendation solutions in the industry still face numerous problems with the knowledge input from large models: insufficient automation, difficulty processing multi-source knowledge, and insufficient generalization. Alternatively, the input to large models is unstructured parameter information, making it difficult for large models to extract accurate value ranges or tuning recommendations. Therefore, the input of large model knowledge in current industry solutions also leads to poor parameter recommendation results and low efficiency.

[0083] Based on this, in the solution provided by this application, a structural parameter library is extracted from heterogeneous knowledge from multiple sources based on a hierarchical prompt chain, and the complex extraction task is decomposed into multiple levels of prompt chains. Combined with the few-sample learning technology, the large model is gradually guided to complete the goals of knowledge organization, knowledge extraction, knowledge integration, etc., thereby improving the task-solving ability of the large model and the generalization of knowledge extraction. This ensures that the knowledge information of the large model input during parameter recommendation can improve the effect and efficiency of the recommendation. In addition, this method of constructing a knowledge parameter library has a high degree of automation and can process parameter knowledge in various fields, ensuring the generalization of the solution.

[0084] The solution provided in this application can be applied to Figure 1 In the illustrated computer system, the computer system is a client / server (C / S) architecture. Figure 1 As shown, the computer system includes a client 101 and a server 102 .

[0085] Figure 1 The application scenarios of the illustrated computer system may include parameter tuning, intelligent diagnosis and operation and maintenance or other scenarios, which are not limited in the present application embodiment. Figure 1 The application scenario of the illustrated computer system is described by taking parameter tuning as an example.

[0086] In the parameter tuning scenario, parameter tuning software is deployed in the server 102 .

[0087] The parameter tuning software in server 102 supports local or cross-device parameter tuning. The parameter tuning software is used to tune the parameters of the operating system and its applications, optimizing the performance of the operating system and its applications. Server 102 can be a server, data center management device, or other product form factors.

[0088] The client 101 is used to provide a human-computer interaction method, and the user initiates commands and requests for tuning tasks through the client 101. After receiving the commands and requests for tuning tasks, the client 101 forwards them to the server 102 to start the parameter tuning process. The server 102 executes the tuning task through the parameter tuning software deployed therein, and optimizes the performance of the object to be tuned corresponding to the tuning task. Specifically, the server 102 implements the solution provided in this application to achieve intelligent parameter tuning. The specific implementation is referred to the following method embodiment section and is not described in detail here.

[0089] Further, Figure 1 The illustrated computer system may be implemented in a single physical device, or Figure 1 The illustrated computer system can be deployed on multiple physical devices, which is not limited in the embodiments of the present application.

[0090] For example, when Figure 1 The illustrated computer system is used in a parameter tuning scenario. Figure 1 The internal architecture of the computer system can be shown as follows Figure 2 As shown. Figure 2 As shown, the client 101 includes hardware resources, including a central processing unit (CPU), input and output (IO), network, and memory. Based on the hardware resources, an operating system (such as a Linux system, an OpenEuler system, or others) is installed. Various application software, such as databases, big data, distributed storage, virtualization, etc., are running on the operating system. The client 101 also provides a tuning engine interface, through which the user initiates a tuning request. After receiving the tuning request, the tuning engine interface initiates the tuning process to the server 102. The parameter tuning software in the server 102 executes the parameter tuning process to optimize the performance of the operating system deployed in the client 101 and the application software running on it.

[0091] certainly, Figure 2 This is only an example of an application scenario of the present application solution and does not constitute a specific limitation.

[0092] Please refer to Figure 3 , is a schematic diagram of the structure of a computing device provided in this application. The computing device is used to execute Figure 1 or Figure 2 The function of the server 102, or the function of the application software deployed in the server 102.

[0093] like Figure 3As shown, the computing device may include a processor 3010 , a bus 3020 , a memory 3030 , and a communication interface 3040 . The processor 3010 , the memory 3030 , and the communication interface 3040 are connected via the bus 3020 .

[0094] It should be understood that in this embodiment, the processor 3010 may be a CPU, but may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0095] The processor 3010 may also be a GPU, an NPU, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the program of the present application.

[0096] The communication interface 3040 is used to implement communication between the computing device and external devices or components.

[0097] The bus 3020 is used to transmit information between the above components (such as the processor 3010 and the memory 3030). In addition to the data bus, the bus 3020 may also include a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 3 In the figure, various buses are labeled as bus 3020. Bus 3020 can be a peripheral component interconnect express (PCIe) bus, an extended industry standard architecture (EISA) bus, a unified bus (Ubus or UB), a compute express link (CXL), a cache coherent interconnect for accelerators (CCIX), etc. It is worth noting that Figure 3 In the example, a computing device including one processor 3010 and one memory 3030 is taken. Here, the processor 3010 and the memory 3030 are respectively used to indicate a type of device or equipment. In a specific embodiment, the number of each type of device or equipment can be determined according to business requirements.

[0098] Memory 3030 may be a volatile memory pool or a nonvolatile memory pool, or may include both volatile and nonvolatile memory. Nonvolatile memory may be read-only memory (ROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0099] It is understood that the structure shown in the embodiment of the present application does not constitute a specific limitation on the computing device. Figure 3 More or fewer components may be shown, two or more components may be combined, or the illustrations may have different configurations of components. Figure 3 The various components shown in the figure can be implemented in hardware, software, or a combination of hardware and software, including one or more signal processing or application-specific integrated circuits. The methods in the following embodiments can all be implemented in a computing device having the above hardware structure.

[0100] The parameter recommendation method provided in this application will be described in detail below with reference to the accompanying drawings.

[0101] On the one hand, the embodiment of the present application provides a parameter recommendation method. Figure 4 This is a flow chart of the parameter recommendation method provided by this application. This method can be applied to a computing device, or software with parameter recommendation function deployed in a computing device. The following embodiments are described by taking a computing device executing this solution as an example, which does not constitute a specific limitation. The computing device can be Figure 1 or Figure 2 The server 102 is shown in FIG.

[0102] like Figure 4 As shown, the parameter recommendation method provided in the embodiment of the present application may include the following steps:

[0103] S401: Obtain a tuning task, where the tuning task includes an object to be tuned and an optimization target.

[0104] Specifically, when a user wishes to perform parameter tuning or diagnostic maintenance on an object to optimize its performance, the user initiates a tuning task through a human-computer interaction interface. The human-computer interaction interface can be a client, an interface, a command, or other form, which is not limited in the present embodiment.

[0105] The object to be tuned is the object to be tuned. It can be an operating system, application software, or an application program. The object to be tuned is represented in the tuning task as information. This identification information uniquely identifies a particular object to be tuned.

[0106] The optimization target is the expected value for parameter tuning. It can be the optimized relative value of the performance indicator or the absolute value of the performance indicator. It can be configured according to actual needs and is not limited.

[0107] In one possible implementation, the tuning task may also include load description information, which indicates the workload of the object being tuned. The load description information can serve as a model prompt, enabling the model to obtain workload information about the object being tuned. For example, the load description information may indicate the specific workload content, stress testing method, or other aspects of the object being tuned. The present embodiments do not limit the content or format of the load description information.

[0108] S402: Collect and analyze the performance data of the object to be adjusted to obtain a performance report of the object to be adjusted.

[0109] The performance report is used to describe the current performance of the object to be adjusted.

[0110] Specifically, in S402, the performance data of the object to be adjusted can be collected through the data collection interface. The content of the collected performance data can be configured according to actual needs, and this embodiment of the application does not limit this.

[0111] In one possible implementation, the performance data collected in S402 may be based on a pre-configured general performance data set, collecting the current values ​​of each performance data in the set. In actual applications, the content of the general performance data set may be configured according to actual needs, and this embodiment of the application is not limited thereto.

[0112] In another possible implementation, the association between the performance indicators and the performance data to be collected can be preconfigured. The performance data associated with a performance indicator is the performance data that affects the performance indicator. The performance data collected in S402 can be performance data associated with the performance indicator involved in the optimization target.

[0113] In one possible implementation, the performance report may include bottleneck information. Of course, the performance report may also include other performance description information. The present embodiment of the application does not specifically limit the content of the performance report. Any information that can reflect the performance problem of the object to be adjusted may be included in the performance report.

[0114] Among them, the performance bottleneck is the key resource or link that limits the overall performance from reaching the optimal level during the operation of the object to be adjusted. Even if the other parts have sufficient capabilities, the overall throughput, response speed or efficiency are still subject to the weakest point of the performance bottleneck. Performance bottlenecks can include bottlenecks in multiple dimensions, which are not limited in the embodiments of this application. For example, performance bottlenecks can include CPU bottlenecks, disk I / O bottlenecks, memory bottlenecks, network bottlenecks or others.

[0115] In one possible implementation, after collecting performance data, a large model may be used to analyze the performance data and obtain a performance report of the object to be adjusted.

[0116] In another possible implementation, after collecting performance data, a clustering algorithm may be used to obtain bottleneck information of the object to be tuned, and a large model may be used to analyze the performance data to obtain a performance report that does not include bottleneck information.

[0117] In another possible implementation, the performance report may also include workload information, which indicates the current workload of the target object. The workload of the target object reflects the busyness, pressure intensity, business throughput, resource consumption efficiency, and resource shortage of the tasks being processed by the target object. The load of the target object is reflected in the form of performance indicators, and the specific content can be configured according to actual needs and is not limited. For example, the workload can be the workload currently being processed by the target object, or the resource usage.

[0118] For example, the workload of an operating system may include: CPU load, memory load, disk I / O load, network I / O load, etc. The workload of an application refers to the resources consumed and the pressure endured by the application in order to process user requests or execute tasks. For example, the workload of an application may include: computing load, storage load, I / O load, or other loads. However, the embodiments of this application are merely illustrative and do not limit the specific content of the workload.

[0119] In a possible implementation, in S402 , workload information in the performance report is obtained by analyzing performance data.

[0120] In another possible implementation, the tuning task acquired in S401 includes load description information. In S402 , the load description information included in the tuning task can be used as a prompt to analyze performance data and obtain workload information in the performance report.

[0121] S403: Based on the performance report, search the parameter knowledge base for the object to be adjusted to determine a parameter set.

[0122] Specifically, in S403, a parameter knowledge base for the current object to be adjusted is first obtained. The parameter knowledge base includes knowledge information for adjusting the configurable parameters of the current object to be adjusted. The content and form of the parameter knowledge base are not specifically limited in this embodiment of the application.

[0123] The knowledge information of a parameter may include parameter identification, parameter description information, etc., without limitation.

[0124] Specifically, when executing S403, first search to determine whether there is a parameter knowledge base for the current object to be adjusted. If so, load the parameter knowledge base and then execute S403. If there is no parameter knowledge base for the current object to be adjusted, then construct the parameter knowledge base. The embodiment of the present application does not limit the method for constructing the parameter knowledge base.

[0125] In one possible implementation, the parameter knowledge base is a structured parameter base extracted from heterogeneous knowledge from multiple sources based on a hierarchical prompt chain to achieve high efficiency and better parameter recommendation. The construction method of the parameter knowledge base is not limited in the embodiment of this application. For example, the parameter knowledge base can be constructed as follows Figure 7 The schematic construction method flow is for illustrative purposes only and is not intended to be limiting.

[0126] In another possible implementation, the parameter knowledge base may be an open-source parameter knowledge base for the current object to be adjusted. The open-source parameter knowledge base may be structured or unstructured. It may be multi-source or single-source. This application does not specifically limit any of these.

[0127] In another possible implementation, the parameter knowledge base may be directly extracted from parameter knowledge from a single source or multiple sources based on a large model.

[0128] Furthermore, in S403, the content of the performance report and the parameter knowledge base can be converted into vector form, and then searched to obtain a parameter set. The parameter set can be the parameters and related knowledge in the parameter knowledge base that match the performance report. When the search method is different, the matching method is different.

[0129] Optionally, when performing the search in S403 , a single search method or a mixed search method combining multiple search methods may be used.

[0130] In one possible implementation, S403 can use semantic retrieval to search based on the various contents in the performance report to determine the similarity (such as cosine similarity) between the contents in the performance report and the knowledge in the parameter knowledge base. The parameter knowledge fragments that meet the similarity conditions are considered to be matching knowledge entries and are determined as the final parameter set.

[0131] The conditions satisfied by the similarity can be configured according to actual needs and are not limited. For example, the condition can be that the similarity with the content of the performance report is greater than or equal to a similarity threshold. The value of the similarity threshold can be configured according to actual needs.

[0132] In another possible implementation, in S403, a keyword search method may be used based on the content in the performance report. Keywords in the performance report are extracted, and knowledge items containing the keywords in the parameter knowledge base are determined to be matching knowledge items and determined as the final parameter set.

[0133] In another possible implementation, S403 can be based on the various contents of the performance report using semantic and keyword search methods, i.e., hybrid search. The semantics and keywords in the performance report are extracted, and the parameter knowledge base is determined to have a semantic similarity with the performance report that meets the conditions, and / or knowledge items containing the keywords are considered matching knowledge items and are determined as the final parameter set.

[0134] Furthermore, the parameter set determined in S403 is an optimal parameter set for the current state of the object to be adjusted, and includes one or more configurable parameters that affect the above-mentioned optimization goal.

[0135] Furthermore, the parameter set may also include value ranges of one or more configurable parameters that affect the above-mentioned optimization objectives.

[0136] In the solution provided by the present application, when a tuning task is obtained, the performance data of the object to be tuned is collected and analyzed to generate a performance report reflecting the performance of the object to be tuned. The parameter knowledge base for the object to be tuned is then searched with reference to the performance report to determine the parameter set for parameter recommendation, replacing the current solution of parameter recommendation using the full input model. Since the performance report of the object to be tuned is combined in the process of determining the parameter set, the configurable parameters in the determined parameter set are strongly correlated with the current performance of the object to be tuned, and the optimal tuning parameters under different performance states can be obtained, so that the quality of parameter recommendation is high and the effect is good.

[0137] Furthermore, in the above Figure 4 In the illustrated parameter recommendation process, S403 derives the optimal tuning parameter set for the current workload based on workload, bottleneck information, and a parameter knowledge base, thus implementing load-aware parameter recommendations. Load-aware RAG technology addresses the inadequate use of domain knowledge in existing solutions. By incorporating knowledge about the current system workload and bottlenecks, it enhances the interpretability of parameter recommendations and accurately recommends performance-related parameters.

[0138] Furthermore, the hybrid retrieval strategy proposed in the above S403, through a hybrid retrieval strategy consisting of semantic retrieval and keyword retrieval, takes advantage of the advantages of different retrieval technologies to achieve accurate screening and semantic expansion of parameter knowledge, improve the accuracy and relevance of retrieval results, and filter out weakly relevant parameter knowledge.

[0139] In a possible implementation, after S403 , the parameter set may be output for user viewing or further operation.

[0140] Furthermore, when Figure 4 The parameter recommendation method shown in the figure is used in parameter tuning scenarios, such as Figure 5 As shown, the method may further include the process of S404.

[0141] S404: Adjust the values ​​of the configurable parameters in the parameter set until a termination condition is reached.

[0142] Specifically, S404 iteratively adjusts the values ​​of the configurable parameters in the parameter set determined in S403 until the termination condition is reached, completing the tuning task. Because the parameter set determined in S403 is the optimal parameter set for the current state of the object to be tuned, ineffective adjustments in S404 are significantly reduced, improving the efficiency and effectiveness of S404.

[0143] The process of adjusting the value of the configurable parameter when executing S404 can be configured according to actual needs, and the embodiment of the present application is not limited to this.

[0144] For example, automatic tuning in S404 works by optimizing configuration parameters and performance metrics of the target object based on the recommended parameter set. It then iterates using the parameter search algorithm in the AI ​​engine (e.g., Bayesian optimization, grid search, etc.) to ultimately determine the optimal parameter configuration based on performance feedback. This parameter configuration serves as the final output of the parameter tuning system.

[0145] In one possible implementation, the parameter set determined in S403 also includes the value ranges of one or more configurable parameters that affect the optimization objective. When adjusting the value of a configurable parameter in S404, the adjustment can be made within the value ranges included in the parameter set, effectively narrowing the adjustment range and improving efficiency.

[0146] In another possible implementation, when adjusting the value of the configurable parameter in S404, the adjustment may be made from preset optional values ​​of the configurable parameter. The optional value may be an empirical value or other values, which are not limited.

[0147] Furthermore, each time the value of the configurable parameter is adjusted in S404, it is determined whether a termination condition is met. If the termination condition is met, the adjustment is terminated.

[0148] The termination condition includes that the performance of the object to be adjusted meets the optimization target. In S404, each time the value of the configurable parameter is adjusted, the performance data of the object to be adjusted is collected to determine whether the performance of the object to be adjusted after the adjustment meets the optimization target.

[0149] Of course, the termination condition may also include other content. For example, the number of iterations is greater than or equal to a threshold. Satisfying the termination condition may mean satisfying any one of the termination conditions, or all of them. The threshold may be configured according to actual needs and is not limited in this embodiment of the present application.

[0150] The parameter recommendation process described in S403 above is described below by way of example. Figure 6 The principle of the parameter recommendation process is illustrated. Figure 6 The illustrated process is only an example of the implementation process of S403 and does not constitute a specific limitation. Figure 6 As shown, the parameter recommendation process includes the following steps 1 to 3:

[0151] Step 1: Parameter knowledge vectorization and index construction.

[0152] In step 1, an embedding model (such as an embedding model) is used to convert the parameter knowledge information in the parameter knowledge base of the object to be adjusted into a vector matrix as a knowledge base vector.

[0153] For example, the process of vector conversion can be expressed as:

[0154] ;

[0155] .

[0156] in, Represents the parameter knowledge base for the object to be adjusted, represents the embedding model used to transform parameter knowledge information into a vector, A vector matrix representing parameter knowledge information (i.e., a knowledge base vector). For example, the parameter knowledge information can be converted into a 1536-dimensional vector representation using the text-embedding-ada-002 model provided by OpenAI.

[0157] Afterwards, the vectors obtained by parameter knowledge information conversion are used to construct an index, and the vector data in the vector matrix is ​​written into the vector database. In step 1, vector representation and index construction are used to achieve efficient storage and fast retrieval of parameter information.

[0158] Step 2: Load-aware retrieval enhancement.

[0159] In step 2, the load information and bottleneck information in the performance report are used for mixed retrieval to filter out irrelevant parameters.

[0160] Specifically, such as Figure 6 As shown in the figure, the text describing the load information and bottleneck information (referred to as load bottleneck information) is vectorized to obtain a query vector. Semantic similarity retrieval is then performed. Specifically, the cosine similarity between the query vector and the knowledge base vector is calculated, and content with high semantic similarity is returned as the retrieval result (essentially, parameter knowledge fragments).

[0161] For example, the calculation process of semantic similarity satisfies the following expression:

[0162] ;

[0163] .

[0164] in, Descriptive text indicating load bottleneck information. The load bottleneck information description text is represented by a vector representation generated by the embedding model; is the vector representation of each parameter knowledge information, is the cosine similarity between the query vector and the knowledge base vector.

[0165] At the same time, if Figure 6As shown, a large model is used to extract keywords and extract tuning keywords from workload bottleneck information. Next, keyword search is performed. For example, a keyword search algorithm such as BM25 is used for precise screening to obtain search results (essentially parameter knowledge fragments). Using the BM25 algorithm as an example, the process of extracting tuning keywords from workload bottleneck information satisfies the following expression:

[0166] ;

[0167] .

[0168] in, Descriptive text indicating load bottleneck information. A set of keywords representing workload bottleneck information generated by a large model. Represents the calculation keyword set and parameter knowledge base text correlation. is a single keyword in the set, is the keyword The pseudo-document word frequency, Indicates keywords In the text The word frequency in Indicates the length of the parameter knowledge base text, represents the average length of all parameter knowledge information in the parameter knowledge base, and is an adjustable parameter. Through this process, the system can effectively evaluate the relevance between query keywords and parameter knowledge information.

[0169] Afterwards, the parameter knowledge obtained by semantic similarity search and keyword search is subjected to parameter knowledge screening as the search result. The parameter knowledge screening can adopt the top-k method, and the value of k can be configured according to actual needs.

[0170] In step 2, through load-aware retrieval enhancement technology and the above-mentioned hybrid retrieval strategy, it is possible to ensure that the retrieval results are highly relevant to the current system status, thereby improving retrieval efficiency and accuracy.

[0171] Step 3: Prompt generation and parameter recommendation.

[0172] In step 3, the parameter knowledge fragments retrieved in step 2 are merged with the performance report to form a complete prompt, which is input into the large model to obtain the parameter recommendation result, that is, the parameter set obtained in S403.

[0173] The parameter set may also include an initial parameter value and an adjustment range.

[0174] Figure 6The illustrated process enhances the tuning reasoning capabilities of large models through retrieval-augmented generation (RAG), generates an optimal parameter set for the current state (load and bottlenecks), and provides recommended initial parameter values ​​and adjustment ranges, reducing parameter dimensions and the number of subsequent tuning iterations.

[0175] On the other hand, the present invention provides a method for constructing a parameter knowledge base. Figure 7 This is a flow chart of the method for building a parameter knowledge base provided by this application. This method can be applied to a computing device, or software with parameter recommendation function deployed in a computing device. The following embodiments are described by taking a computing device executing this solution as an example, which does not constitute a specific limitation. The computing device can be Figure 1 or Figure 2 The server 102 is shown in FIG.

[0176] Figure 7 The schematic method of constructing parameter knowledge base can be used with Figure 4 The illustrated parameter recommendation methods may be used in combination or individually, and the embodiments of the present application are not limited thereto. Figure 7 The illustrated method for constructing a parameter knowledge base can be executed when executing a user-initiated tuning task for performing parameter tuning on an object to be tuned, by searching and determining that no parameter knowledge base for the object to be tuned exists; or, it can also be executed when resources are sufficient, or at other times. The embodiment of the present application does not specifically limit the timing of executing the method for constructing a parameter knowledge base.

[0177] Furthermore, the construction method of parameter knowledge base for different objects is the same, as follows Figure 7 In this article, only the process of building a parameter knowledge base for the object to be adjusted is described as an example, and the others are not described in detail.

[0178] like Figure 7 As shown, the method for constructing a parameter knowledge base may include:

[0179] S701: Obtain parameter information for an object to be adjusted from multiple sources.

[0180] The objects to be adjusted described in the method for constructing the parameter knowledge base are the objects involved in constructing the parameter knowledge base, that is, Figure 6 The process of constructing a parameter knowledge base for the object to be adjusted. The object to be adjusted can be the aforementioned Figure 4 The object to be adjusted described in , or it can also be other objects to be adjusted, without limitation.

[0181] Specifically, in S701, parameter information for the object to be adjusted can be collected from multiple different sources. For example, the sources of the parameter information from different sources may include but are not limited to: web pages, structured documents, source code, technical documents, etc. Of course, the embodiment of the present application does not limit the source of the parameter information.

[0182] For example, in S701, the parameter information obtained from multiple sources can be converted into a natural language text format to facilitate understanding by the large model. The embodiment of the present application does not limit the conversion process.

[0183] Optionally, in S701 , the parameter information may be divided into sliding window documents to facilitate the large model to extract parameter knowledge therefrom.

[0184] S702: Based on the hierarchical prompt chain, LLM is used to extract parameter knowledge from various sources from the parameter information.

[0185] Specifically, in S702 , the parameter information from multiple sources is perceived and understood respectively through the LLM technology, and then the content useful for parameter tuning is extracted from the parameter information from each source as the parameter knowledge from each source.

[0186] Furthermore, the LLM technology used in S702 is based on hierarchical prompt chaining. Hierarchical prompt chaining is an artificial intelligence engineering method that breaks down complex tasks into multiple layers of subtasks and executes them in series through ordered prompts. The process of extracting parameter knowledge from parameter information based on hierarchical prompt chaining with LLM technology in this embodiment of the application will not be further described.

[0187] Furthermore, when executing S702 , a prompt template may be designed through the manually labeled sample pool so that the LLM extracts parameter knowledge from various sources from the parameter information according to the structure of the sample pool.

[0188] Furthermore, in S702 , the knowledge stored in the large model pre-training can be used as a knowledge supplement and determined as a source of parameter knowledge, thereby further enriching the source of the extracted data.

[0189] For example, the parameter knowledge extracted in S702 may include: structured parameter knowledge including specifications of field parameters, how to set and validate and query, whether restart is required to take effect, and other information, as well as deeper tuning experience knowledge.

[0190] S703: Fuse parameter knowledge from various sources to obtain a parameter knowledge base.

[0191] Specifically, fusion refers to aggregating parameter knowledge from various sources into a single parameter knowledge base. During the fusion process, non-conflicting parameter knowledge is directly recorded in the parameter knowledge base. Conflicting parameter knowledge can be handled using a fusion strategy. This results in a parameter knowledge base. The content of the fusion strategy can be configured based on actual needs and is not limited in this embodiment of the present application.

[0192] In one possible implementation, the fusion strategy can be a confidence strategy, which selects parameter knowledge with high confidence and records it in the parameter knowledge base. Confidence is used to describe the credibility of parameter knowledge. In practical applications, the process of determining the confidence of parameter knowledge can be configured according to actual needs and is not limited by the embodiments of this application.

[0193] Exemplarily, an embodiment of the present application illustrates a process for executing S703 based on confidence, including: using a large model to determine the confidence information of fields in parameter knowledge from various sources, the confidence information is used to describe the credibility of the field; based on the confidence information, constructing fusion prompt information, the fusion prompt information is used to indicate the credibility of each field in different sources; according to the fusion prompt information, using the large model to select the field with the highest confidence, and fusing the parameter knowledge from various sources to obtain a parameter knowledge base.

[0194] The confidence information can be used to describe the reliability, completeness or consistency of the field.

[0195] For example, the field confidence can be determined to satisfy the following expression:

[0196] .

[0197] in, represents the confidence quantization result of the jth field of the i-th knowledge source; Represents the structured parameter knowledge of the i-th knowledge source; Indicates the source of knowledge The reliability of field j in Indicates its completeness, Indicates its consistency. is a weight coefficient that can be configured based on actual needs. The reliability, completeness, and consistency of a field can be uniformly scored by the large model, and the quantitative results of the field confidence assessment are ultimately calculated.

[0198] Through the method for constructing a parameter knowledge base provided in this application, a structural parameter library is extracted from heterogeneous knowledge from multiple sources based on a hierarchical prompt chain, and complex extraction tasks are decomposed into prompt chains at multiple levels. Combined with the few-sample learning technology, the large model is gradually guided to complete the goals of knowledge organization, knowledge extraction, knowledge integration, etc., thereby improving the task-solving ability of the large model and the generalization of knowledge extraction.

[0199] Furthermore, the confidence perception mechanism proposed in this method of constructing a parameter knowledge base is conducive to the large model determining the priority of each field under each source, improving the accuracy of the parameter knowledge base results, and solving the problems of multi-source data conflicts and insufficient field integrity and low overall accuracy faced by existing solutions.

[0200] The following examples illustrate the above Figure 7 The process of building a parameter knowledge base is illustrated with an example. Figure 8 The principle of the method for constructing a parameter knowledge base is illustrated. Figure 8 The process of indicating is just Figure 7 This is just an example of the solution and does not constitute a specific limitation. Figure 8 As shown, the process of constructing the parameter knowledge base includes the following steps a to c:

[0201] Step a: Multi-source heterogeneous data collection and preprocessing.

[0202] In step a, data collection is used to extract multi-source parameter information (relevant information for tuning) from different sources (e.g., web pages, structured documents, source code, etc.), and then heterogeneous document parsing is performed.

[0203] Before parsing heterogeneous documents, they can be converted into natural language text format through data conversion. The data format conversion process can be expressed as:

[0204] ;

[0205] ;

[0206] .

[0207] in, 、 、 Respectively represent web pages, PDF, Excel and other documents and source code content, Represents natural language text, conversion function 、 、 Convert heterogeneous source content into natural language text for easy subsequent processing. For example, the conversion function implementation includes: obtaining HTML pages from web pages using the BeautifulSoup library and cleaning the web content using the Readability tool. PDF documents are parsed using the PyMuPDF library. Excel documents are read from rows and columns using the Pandas library, and similarities between rows and columns are calculated, followed by order analysis and precise extraction. Parameter information in source code is extracted using an abstract syntax tree (AST) and regular expressions.

[0208] Then, the long document is segmented into sliding window documents to obtain multi-source parameter text blocks. The specific method of the sliding window document segmentation is not limited in the present embodiment. The sliding window document segmentation can be expressed as follows:

[0209] .

[0210] in, Represents a long text converted from multi-source heterogeneous documents into a natural language format. represents the overlapping ratio of sliding windows, Represents the set of text blocks after segmentation. During the sliding window segmentation, the text overlap method is used to prevent context loss and form parameter knowledge text blocks that are easy to process for large models.

[0211] Step b: Structured parameter extraction and knowledge supplementation.

[0212] In step b, we first construct samples and design a multi-level prompt chain. Specifically, we design a prompt template for few-shot learning using a pool of manually annotated samples. This multi-level design includes a prompt chain for extracting key parameter fields, a prompt chain for analyzing field confidence based on historical data, and a prompt chain for multi-source data fusion with a confidence-priority priority.

[0213] Then, the large model is called to extract the structured parameter knowledge information from the multi-source parameter text block as the multi-source structured parameter information. This process can be represented as With input text block and sample examples Combine to form prompt input ; and As the input of the big model, the big model generates structural parameter information. This process can be expressed by the following expression:

[0214] ;

[0215] .

[0216] in, Indicates the operation of combining the prompt template with the text block and sample. Extract is a function that represents the large model LLM in the prompt input Under the guidance of .

[0217] Furthermore, the knowledge existing in the pre-training of large models can be used as a knowledge supplement for structured parameter information to further enrich the extracted data sources.

[0218] The process of knowledge supplementation can be expressed as:

[0219] .

[0220] in, It is parameter knowledge information extracted from multi-source parameter text blocks. Parameter knowledge information during the pre-training process of large models.

[0221] Step c: Confidence-prioritized multi-source knowledge fusion.

[0222] In step c, a field confidence analysis is first performed to evaluate the confidence of each field in the parameter knowledge information extracted from the multi-source parameter text block in step b, so as to quantify the reliability and completeness of each data source field.

[0223] Based on the evaluation results, the system then designs large-scale model prompts for the knowledge fusion process according to the field confidence level, ensuring that subsequent processing is more inclined to utilize this high-quality data. For example, the designed large-scale model prompts can indicate the confidence level of which field from which source, so that the large-scale model can select high-quality parameter knowledge for fusion based on the prompts. This embodiment of the application does not limit the process of designing large-scale model prompts.

[0224] Finally, the large model performs confidence-aware knowledge fusion through prompts, checks and fuses multi-source data to form a structured parameter knowledge base.

[0225] In steps b and c above, a carefully designed multi-level large model prompt chain ensures that the extraction results are accurate and complete. At this point, the parameter knowledge base is constructed.

[0226] On the other hand, the embodiment of the present application provides another parameter recommendation method. Figure 9 This is a flow chart of the parameter recommendation method provided by this application. This method can be applied to a computing device, or software with parameter recommendation function deployed in a computing device. The following embodiments are described by taking a computing device executing this solution as an example, which does not constitute a specific limitation. The computing device can be Figure 1 or Figure 2 The server 102 is shown in FIG.

[0227] like Figure 9 As shown, the parameter recommendation method provided in the embodiment of the present application may include the following steps:

[0228] S901: Obtain a tuning task, where the tuning task includes an object to be tuned and an optimization target.

[0229] The implementation of S901 may refer to the aforementioned process of S401 and will not be repeated here.

[0230] S902: Search the parameter knowledge base for the object to be adjusted to determine a parameter set. The parameter knowledge base is a structural parameter base extracted from heterogeneous knowledge sources based on a hierarchical prompt chain.

[0231] Through S902, high efficiency and better effect parameter recommendation are achieved. The specific construction process of the parameter knowledge base is not limited in this embodiment of the application. All structural parameter libraries extracted from multi-source heterogeneous knowledge based on the hierarchical prompt chain belong to the parameter knowledge base described in S902. For example, the parameter knowledge base can be constructed according to the above Figure 7 The schematic construction method flow is for illustrative purposes only and is not intended to be limiting.

[0232] The search process in S902 may be a full input search, or may be performed according to the search process described in S403. The embodiment of the present application does not limit the search process in S902.

[0233] Furthermore, the parameter set may also include value ranges of one or more configurable parameters that affect the above-mentioned optimization objectives.

[0234] Furthermore, if Figure 9 The illustrated parameter recommendation method is applied to the parameter tuning scenario. After S902, the value of the configurable parameters in the parameter set can also be adjusted until the termination condition is reached. The specific implementation of this process can refer to the aforementioned S404 process and will not be repeated here.

[0235] This parameter recommendation method extracts a structured parameter library from heterogeneous knowledge from multiple sources using a hierarchical prompt chain. This method decomposes the complex extraction task into multiple layers of prompt chains. Incorporating few-shot learning techniques, it gradually guides the large model through knowledge organization, extraction, and integration, improving both its task-solving capabilities and the generalizability of knowledge extraction. This ensures that the knowledge information input into the large model during the parameter recommendation process improves both the effectiveness and efficiency of tuning. Furthermore, this method for constructing the knowledge parameter library is highly automated and can handle parameter knowledge from a variety of fields, ensuring the generalizability of the solution.

[0236] Figure 10The specific application scenarios of the parameter recommendation method provided in this application are illustrated. Figure 10 The illustrated application scenario is the parameter recommendation method provided by this application, which is applied in the parameter tuning scenario. Figure 10 As shown, the application scenario includes the client and the server. The internal architecture of the client is Figure 2 Based on the schematic, it also includes data collection plug-ins, load execution plug-ins, and indicator collection plug-ins. The server includes a data collection module, a load perception module, a parameter recommendation module, a parameter optimization module, and a knowledge base construction module. Figure 10 The working principle of the illustrated application scenario includes:

[0237] The user initiates a tuning task on the client, and the tuning engine interface retrieves the task. The tuning engine interface transmits the tuning task to the server, initiating the tuning process. Upon receiving the tuning task, the server invokes the data collection plug-in through the data collection module to collect performance data from the operating system deployed on the client and transmit it to the load sensing module. The load sensing module analyzes and perceives the performance data, generates a performance report, and passes it to the parameter recommendation module. The knowledge base construction module extracts parameter knowledge from multiple source tuning documents and constructs a parameter knowledge base for tuning. This knowledge base participates in the parameter recommendation process as domain knowledge and is passed to the parameter recommendation module. Based on the performance report and the parameter knowledge base for tuning, the parameter recommendation module recommends the optimal parameter set for the current system state. Finally, based on the optimal parameter set, the parameter optimization module executes the workload through multiple iterations of the workload through the load execution plug-in. Using metrics collected by the metric collection plug-in, the module determines whether the current performance meets the optimization target, explores the parameter configuration space, and generates the final parameter configuration using algorithms such as Bayesian optimization.

[0238] Further, Figure 10 The server shown in the figure can be implemented in software as follows Figure 11 As shown, the knowledge base construction module can deploy multi-source knowledge extraction technology based on hierarchical prompt chains and confidence-aware knowledge fusion technology. The load sensing module can deploy application-level load sensing technology to perceive application load information, and system-level load sensing technology can also be deployed to perceive operating system-level load information. The parameter recommendation module can deploy load-aware retrieval enhancement technology and hybrid retrieval strategies. The parameter optimization module can deploy Bayesian optimization algorithms and reinforcement learning algorithms.

[0239] The following describes the specific implementation of the solution of the present application in a parameter tuning scenario through specific examples.

[0240] In this example, the parameter tuning software in the cloud service scenario is used as an example. Figure 7The schematic scheme builds a parameter knowledge base, using this application Figure 4 This example demonstrates parameter recommendations and optimization. This parameter optimization software runs on a virtual machine deployed with a local large language model and an embedded model, enabling parameter optimization for applications in cloud service scenarios. The large language model leverages its powerful natural language processing capabilities to provide parameter extraction and optimization recommendations, while the embedded model vectorizes parameter information for efficient knowledge retrieval. This example can significantly improve the efficiency and effectiveness of performance optimization for cloud service applications.

[0241] In the cloud service parameter tuning software application, this example involves the steps of parameter knowledge base construction and parameter tuning. Figure 8 The illustrated process is based on multi-source heterogeneous tuning documents, and through multi-level prompt chains and field confidence-aware knowledge fusion technologies, a parameter knowledge base for tuning tasks is generated. Figure 6 Schematic process,designing load-aware retrieval enhancement technology,using application workload and bottleneck information to identify,important tuning parameters.

[0242] The following example illustrates the solution provided in the application by tuning the parameters of a MySQL database application.

[0243] The first step is pre-tuning. This is performed during the initial preparation phase of a MySQL tuning task and requires building a knowledge base for MySQL tuning parameters. The existing MySQL knowledge base includes official parameter documentation and tuning experience from database forums. This data contains a wealth of knowledge required for parameter tuning, such as parameter descriptions, parameter values, and whether parameters are hot-effective.

[0244] This preparation phase includes the following steps: multi-source heterogeneous data collection and preprocessing, structured parameter extraction and knowledge supplementation, and confidence-prioritized multi-source knowledge fusion. Each step is described below.

[0245] (1) Multi-source heterogeneous data collection and preprocessing.

[0246] The specific process of multi-source heterogeneous data collection and preprocessing can be referred to Figure 8 For example, the process of multi-source heterogeneous data collection and preprocessing can be as follows: Figure 12 As shown, including:

[0247] S1201. Collect parameter documents.

[0248] For example, we collected MySQL official documents in web format, MySQL tuning parameter recommendation websites, and MySQL performance documents in PDF format.

[0249] S1202: Determine whether the parameter document is in a format that supports parsing.

[0250] If yes, execute the operations of S1203 and S1204; if no, end this tuning process.

[0251] S1203. Convert the data format based on the document type into a unified natural language text format.

[0252] For example, parameter-related information can be converted into a natural language text format through tools such as the BeautifulSoup library and the PyMuPDF library.

[0253] S1204. Design a sliding window and context overlapping method to divide the long document information into data and generate parameter knowledge text blocks.

[0254] Long documents are divided based on sliding windows to form MySQL parameter knowledge text blocks for data preprocessing.

[0255] (2) Structured parameter extraction and knowledge supplementation.

[0256] This step is executed after obtaining the parameter knowledge text block. For the specific process, please refer to Figure 8 The process of step b is shown in FIG.

[0257] For example, the specific execution process of the structured parameter extraction and knowledge supplementation steps can be as follows: Figure 13 Shown, including:

[0258] S1301: Check whether there is a labeled structured parameter knowledge sample.

[0259] If it exists, the process of S1302 is executed. If it does not exist, the process of S1303 is skipped.

[0260] S1302. Add the structured parameter knowledge sample into the designed large model prompt chain.

[0261] Through S1302, a hint enhancement based on a few-shot learning method is formed. Thereafter, the process of S1303 is executed.

[0262] S1303. Parse the parameter knowledge text block based on the prompt chain to generate structured knowledge information from multiple sources and store it in JSON format.

[0263] S1304: Based on the extracted parameter names, the domain knowledge existing in the large model pre-training is used as a knowledge supplement to generate structured parameter knowledge from various sources.

[0264] For example, by designing a large language model with a multi-level prompt chain and structured sample templates, the deployed large language model is driven to gradually extract structured parameter information from multi-source parameter text blocks and store it in JSON format. While the parameter knowledge text already contains extensive tuning experience, it still lacks knowledge on how to set, validate, and query MySQL parameters. Therefore, this step supplements the knowledge gained from large model pre-training to fill in the gaps in the open source documentation, facilitating analysis and utilization by tuning tools. The fields of structured parameter information can be shown in Table 1.

[0265] Table 1

[0266]

[0267] (3) Confidence-prioritized multi-source knowledge fusion.

[0268] This step is performed after obtaining the multi-source structural parameter knowledge. The specific process can be referred to Figure 8 The process of step c is shown in FIG.

[0269] For example, the specific execution process of the confidence-priority multi-source knowledge fusion link can be as follows: Figure 14 Shown, including:

[0270] S1401. Divide multi-source structured parameter knowledge according to fields and sources, and format the source and field values ​​in a key-value pair format.

[0271] S1402: Take the key-value pairs as input and use the big model to evaluate the confidence of each data source field.

[0272] Taking a large number of key-value pairs as input, a large model is used to evaluate the confidence of each data source field, quantifying its reliability and completeness. For example, parameter value ranges in official documentation generally have a high confidence level, while parameter tuning recommendations in database forums are also highly confident.

[0273] S1403. Based on the confidence perception results, a large model prompt is designed. The large model checks and fuses multi-source data according to the prompt to generate a structured parameter knowledge base.

[0274] Based on the confidence assessment results, a large-scale knowledge fusion model is designed to prioritize high-confidence data sources. Based on these prompts, the large-scale model examines and integrates structured parameter knowledge from multiple sources (obtained during the structured parameter extraction and knowledge supplementation phases), resolving parameter conflicts between different sources. Ultimately, a structured MySQL parameter knowledge base is formed, ensuring the accuracy and reliability of subsequent tuning recommendations.

[0275] The following describes the specific steps for parameter tuning in this example, including: application load operation and performance data collection, load awareness and bottleneck analysis, parameter knowledge vectorization and index construction, load-aware search enhancement, prompt generation and parameter recommendation, and automated parameter configuration optimization. Each step is described below.

[0276] (1) Application load operation and performance data collection.

[0277] When receiving the start command of the tuning task initiated by the user, the application load operation and performance data collection steps are executed. For the specific process, please refer to Figure 4 The process of S402 is shown in FIG.

[0278] Specifically, the data collection module invokes a data collection plug-in to collect data on the cloud server's resource consumption and performance metrics. For example, collected performance metrics may include CPU utilization, context switches, memory usage, disk I / O utilization, and disk queue length. The collected data in this step is output and used by the load-aware model to analyze the stress load of the current working node.

[0279] (2) Load perception and bottleneck analysis.

[0280] This step is performed after performance metrics are collected. Based on the performance data provided by the data collection module, the pressure on the current working node and resource allocation are analyzed to generate a performance analysis report that includes system bottlenecks. Bottleneck detection can rely on clustering algorithms to analyze and obtain bottleneck information. Bottleneck analysis results (i.e., bottleneck information) can include: CPU bottlenecks, disk I / O bottlenecks, memory bottlenecks, network bottlenecks, etc. Other content in the performance analysis report can be generated by the large model in combination with prompts to provide optimization suggestions for subsequent parameter tuning.

[0281] (3) Parameter knowledge vectorization and index construction.

[0282] This step is performed after the performance analysis report is generated. For the specific process, please refer to the previous Figure 6 Implementation of step 1 in . In this step, since the constructed parameter knowledge base is structured data, the text segmentation process before knowledge embedding can be omitted, and each vector index is guaranteed to contain only the data of a single parameter knowledge, so that the accuracy of subsequent retrieval results can be improved.

[0283] For example, the process of parameter knowledge vectorization and index construction can be as follows Figure 15 Shown, including:

[0284] S1501. Use the embedding model deployed on the machine to convert the parameter knowledge information in the parameter knowledge base into a vector matrix.

[0285] S1502: Build a vector index architecture, write the data in the vector matrix into the vector database, and obtain a parameter knowledge base vector.

[0286] For example, the text in the parameter knowledge base can be converted into a 1536-dimensional vector representation using the text-embedding-ada-002 model and stored in the Faiss vector database to obtain the parameter knowledge base vector. This method enables efficient storage of MySQL parameter knowledge and facilitates subsequent rapid retrieval.

[0287] (4) Load-aware retrieval enhancement.

[0288] This step is performed after the content embedding vector database of the parameter knowledge base is completed. The specific process can refer to the above Figure 6 Implementation of step 2 in . In this step, load-aware search enhancement capabilities are implemented based on MySQL. For example, the process of load-aware search enhancement can be as follows: Figure 16 Shown, including:

[0289] S1601: Determine whether the search operation to be performed is a keyword search or a semantic search.

[0290] If it is a keyword search, execute the operations of S1602 and S1603; if it is a semantic search, execute the operations of S1604 and S1605; if it is a mixed search, execute the operations of S1602 and S1603, and the operations of S1604 and S1605 respectively.

[0291] S1602. The large model combines application types to identify high-frequency tuning keywords in performance analysis reports.

[0292] For example, identifying high-frequency keywords in MySQL performance optimization (for example, index optimization, query optimization, etc.).

[0293] S1603. Based on the keyword search algorithm, the parameter knowledge in the parameter knowledge base is matched, and the matching parameter text block is returned.

[0294] For example, use the BM25 keyword search algorithm to accurately screen relevant parameters.

[0295] S1604: Encode the performance analysis report text of the MySQL server into a query vector.

[0296] For example, the embedding model is used to encode the performance analysis report text into a query vector.

[0297] S1605: Calculate the cosine similarity between the query vector and the parameter knowledge base vector, and return parameter text blocks with high semantic similarity.

[0298] After S1603 and S1605, the process of S1606 is performed.

[0299] S1606. Summarize the parameter text blocks matched by the hybrid search strategy.

[0300] S1606 summarizes the parameter text blocks matched based on the hybrid search strategy to ensure relevant and accurate coverage. For example, deduplication operations or other operations can be performed in S1606.

[0301] Figure 16 The illustrated process ensures that the retrieval results are highly relevant to the current system operation status through a hybrid retrieval strategy. Figure 16 The parameter text block obtained in the illustrated process is a piece of parameter knowledge related to MySQL tuning.

[0302] (5) Prompt generation and parameter recommendation.

[0303] This step is executed after the previous step of retrieval is completed and before the parameter optimization module. The specific process can be referred to the above Figure 6 For example, the process of prompt generation and parameter recommendation can be as follows: Figure 17 As shown, including:

[0304] S1701: Merge the retrieved parameter text block with the original performance analysis report to form complete prompt information.

[0305] Use prompt information as input to the large model to enhance its tuning and reasoning capabilities.

[0306] S1702. The large model generates an optimal set of MySQL parameters for the current state based on the current MySQL load status and historical tuning experience.

[0307] The parameter set may provide specific parameter initial values ​​and adjustment range suggestions.

[0308] After this step, the parameter recommendation process is completed.

[0309] (6) Automation parameter configuration optimization.

[0310] This step is executed after the parameter recommendation is completed to perform automatic tuning. The specific process can refer to the description of S404 above.

[0311] The parameter knowledge base construction method provided in the above example is based on parameter knowledge extraction technology using a hierarchical prompt chain, combined with field confidence-aware multi-source parameter knowledge fusion technology. It leverages the power of large models to automatically extract tuning parameter information from heterogeneous sources. This addresses the problems of traditional knowledge base construction methods, such as low automation and inability to fuse information using heterogeneous knowledge from multiple sources, resulting in poor accuracy in key field extraction and a lack of cross-scenario generalization. This parameter knowledge base construction method significantly improves extraction efficiency across multiple application scenarios compared to existing methods, achieving an 80% accuracy rate for extracted fields. The method's generalizability has been verified in applications such as MySQL and Redis.

[0312] The parameter recommendation method provided in the above example implements load-aware retrieval enhancement technology. It utilizes a hybrid retrieval strategy combining model-embedded semantic retrieval with keyword retrieval, combining performance analysis reports with a parameter knowledge base to generate prompts that drive large-scale model inference. This method recommends optimal parameters for specific scenarios based on application load and bottleneck information. It can also be combined with tuning tools to enable rapid parameter tuning, improving both efficiency and effectiveness. For example, in a MySQL scenario, the parameters recommended by this method improved tuning performance compared to those based on expert experience. MySQL's queries per second (QPS) increased by approximately 12.26%, exceeding the 8.39% improvement achieved using expert experience. This approach addresses the issue of poor recommendation quality caused by the difficulty in integrating domain knowledge and workload into the tuning process.

[0313] Therefore, the solution in this example ultimately improves the efficiency and effectiveness of parameter tuning.

[0314] Furthermore, the solution provided in this application can be applied not only to the field of parameter tuning, but also to other fields, such as intelligent diagnosis and intelligent operation and maintenance.

[0315] For example, in the field of database diagnostics, Figure 7 The illustrated knowledge base construction technology can be used to collect database anomaly information and experience to form an anomaly diagnosis knowledge base. Figure 4 The illustrated parameter recommendation technology can be used in the large-model diagnosis process to output corresponding diagnostic results based on database abnormality information. In this scenario, the object to be adjusted is the system or application to be diagnosed, and the optimization goal is to perform intelligent diagnosis and operation and maintenance. The recommended parameters can be abnormal information or existing problems of the system. The solution provided in this application can transform the process of traditional diagnosis and operation and maintenance that relies on expert experience into a data-driven automated intelligent service. The specific implementation process can refer to the description of the aforementioned method embodiment and will not be repeated here.

[0316] It is understandable that in order to implement the functions in the above embodiments, the electronic device includes hardware structures and / or software modules corresponding to the execution of each function. It should be readily apparent to those skilled in the art that, in combination with the units and method steps of each example described in the embodiments disclosed in this application, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application scenario and design constraints of the technical solution.

[0317] Combined with the above Figures 1 to 17 , describes in detail the parameter tuning method provided by this application, and will be combined with Figures 18 to 20 , describing the apparatus provided by this application. These apparatuses can be used to implement the functions of the computing device in the above method embodiment, and thus can also achieve the beneficial effects possessed by the above method embodiment. In this embodiment, the apparatus can be as follows Figure 3 The processor in the computing device shown in may also be a module (such as a chip) applied to the computing device.

[0318] like Figure 18 As shown, the parameter recommendation device 180 provided in this application includes an acquisition unit 1801 and a parameter recommendation unit 1802.

[0319] The parameter recommendation device 180 is used to implement the above Figure 4 or Figure 5 The method embodiments shown in FIG.

[0320] Exemplarily, the acquisition unit 1801 is used to execute Figure 4 or Figure 5 Step S401. Parameter recommendation unit 1802, for executing Figure 4 or Figure 5 Step S402 or S403.

[0321] Further, such as Figure 19 As shown, the parameter recommendation device 180 may further include an optimization unit 1803. The optimization unit 1803 is used to perform Figure 5 Step S404.

[0322] like Figure 20 As shown, the device 200 for constructing a parameter knowledge base provided by the present application includes an acquisition unit 2001 , an extraction unit 2002 and a fusion unit 2003 .

[0323] The device 200 for constructing a parameter knowledge base is used to implement the above Figure 7 The method embodiments shown in FIG.

[0324] Exemplarily, the acquisition unit 2001 is used to execute Figure 7 Step S701. Extraction unit 2002, for executing Figure 7 Step S702. The fusion unit 2003 is used to perform Figure 7 Step S703.

[0325] It should be understood that the parameter recommendation device 180 and the device 200 for constructing the parameter knowledge base in the embodiment of the present application can be implemented by an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), and the above-mentioned PLD can be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof. It can also be implemented by software. Figure 4 or Figure 7 When the method shown in FIG. 1 is used, its various modules may also be software modules, and the parameter recommendation device 180, the device for constructing a parameter knowledge base 200 and its various modules may also be software modules.

[0326] According to the parameter recommendation device 180 and the device 200 for constructing the parameter knowledge base of the embodiment of the present application, the method described in the embodiment of the present application may be executed correspondingly, and the above and other operations and / or functions of each unit in the parameter recommendation device 180 and the device 200 for constructing the parameter knowledge base are respectively to realize Figure 4 or Figure 7 For the sake of brevity, the corresponding processes of each method in are not repeated here.

[0327] In another aspect, the present application also provides a computer program product comprising instructions. The computer program product may be a software or program product comprising instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes the above-mentioned Figure 4 or Figure 7 The method of indication.

[0328] On the other hand, the embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the above Figure 4 or Figure 7 The method of indication.

[0329] The method steps in this embodiment can be implemented via hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), registers, a hard disk, a removable hard disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC. Alternatively, the ASIC can be located in a computing device. Of course, the processor and storage medium can also exist as discrete components in a computing device.

[0330] The above embodiments can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer programs or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video disc (DVD); or a semiconductor medium, such as a solid-state drive (SSD). The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A parameter recommendation method, characterized in that: The method comprises: Obtaining a tuning task, wherein the tuning task includes an object to be tuned and an optimization target; Collecting and analyzing the performance data of the object to be adjusted to obtain a performance report of the object to be adjusted, wherein the performance report is used to describe the current performance of the object to be adjusted; Based on the performance report, a parameter knowledge base for the object to be adjusted is searched to determine a parameter set, where the parameter set includes one or more configurable parameters that affect the optimization target; the parameter knowledge base includes knowledge information for adjusting the configurable parameters of the object to be adjusted; The method further comprises: Acquiring parameter tuning information for the object to be tuned from heterogeneous knowledge from multiple sources, wherein the parameter tuning information includes parameter knowledge from various sources; Determining confidence information of fields in the parameter knowledge from each source, wherein the confidence information is used to describe the credibility of the fields; The parameter knowledge from the various sources is fused according to the confidence information to obtain the parameter knowledge base.

2. The method according to claim 1, characterized in that The performance report includes bottleneck information and / or workload information. The bottleneck information is used to describe the performance bottleneck of the object to be adjusted, and the workload information is used to describe the workload of the object to be adjusted.

3. The method according to claim 1 or 2, characterized in that The performance report includes workload information, and the tuning task further includes load description information, where the load description information is used to indicate the workload of the object to be tuned; The collecting and analyzing the performance data of the object to be adjusted to obtain a performance report of the object to be adjusted includes: The workload description information is used as a prompt to analyze the performance data to obtain the workload information in the performance report.

4. The method according to claim 1 or 2, characterized in that The retrieving the parameter knowledge base based on the performance report includes: Based on the performance report, the parameter knowledge base is searched using semantic and / or keyword search methods.

5. The method according to claim 1 or 2, characterized in that The parameter knowledge base is a structural parameter base extracted from the multi-source heterogeneous knowledge based on a hierarchical prompt chain.

6. The method according to claim 5, characterized in that The method further comprises: Based on the hierarchical hint chain, a large language model (LLM) is used to extract the parameter knowledge from the various sources from the parameter tuning information.

7. The method according to claim 6, characterized in that The parameter knowledge from various sources is integrated according to the confidence information to obtain the parameter knowledge base, including: Constructing fusion prompt information based on the confidence information, wherein the fusion prompt information is used to indicate the credibility of each field in different sources; According to the fusion prompt information, the field with the highest confidence is selected, and the parameter knowledge from the various sources is fused to obtain the parameter knowledge base.

8. The method according to claim 1 or 2 or 6 or 7, characterized in that The method further comprises: The values ​​of the configurable parameters in the parameter set are adjusted until a termination condition is met, where the termination condition includes that the performance of the object to be adjusted meets the optimization goal.

9. The method according to claim 8, characterized in that The termination condition includes the number of iterations being greater than or equal to a threshold.

10. A parameter recommendation device, characterized in that: The device comprises: An acquisition unit, configured to acquire an optimization task, wherein the optimization task includes an object to be adjusted and an optimization target; a parameter recommendation unit configured to collect and analyze performance data of the object to be adjusted to obtain a performance report of the object to be adjusted, wherein the performance report is used to describe the current performance of the object to be adjusted; based on the performance report, a parameter knowledge base for the object to be adjusted is searched to determine a parameter set, wherein the parameter set includes one or more configurable parameters that affect the optimization objective; the parameter knowledge base includes knowledge information for adjusting the configurable parameters of the object to be adjusted; The parameter knowledge base is constructed by a device for constructing a parameter knowledge base, and the device for constructing a parameter knowledge base includes an acquisition unit and a fusion unit; The acquisition unit of the apparatus for constructing a parameter knowledge base is configured to acquire parameter tuning information for the object to be tuned from heterogeneous knowledge from multiple sources, wherein the parameter tuning information includes parameter knowledge from various sources; The fusion unit is used to determine the confidence information of the fields in the parameter knowledge from the various sources, where the confidence information is used to describe the credibility of the fields; based on the confidence information, the parameter knowledge from the various sources is fused to obtain the parameter knowledge base.

11. A computing device, characterized in that The computing device includes a processor and a memory; The processor is configured to execute instructions stored in the memory, so that the computing device performs the operating steps of the method according to any one of claims 1 to 9.

12. A computer program product comprising instructions, characterized in that When the instructions are executed by a computing device, the computing device is caused to perform the operation steps of the method according to any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a computing device, the computing device performs the operation steps of the method according to any one of claims 1 to 9.

Images