Agent-based computing resource requirement identification and scheduling recommendation method and agent

Abstract

This invention provides a method and agent-based approach for identifying and scheduling computing resource requirements, relating to the field of cloud computing resource management technology. The method includes: receiving a user's fuzzy description of computing resource requirements; parsing the user's requirements using natural language processing (NLP) technology to identify user intent and key entities; proactively guiding the user to complete missing parameters through multi-turn dialogue based on key entities and a built-in knowledge base; calling a resource evaluation model to quantify the parameters and generate a list of computing resource requirements; obtaining a set of candidate resources that meet the list of requirements; calculating multi-dimensional utility scores for each candidate resource; generating multiple recommendation schemes based on different weighting strategies across multiple dimensions, and generating natural language descriptions of recommendation reasons for each scheme; and executing resource allocation based on the user's selection. This invention improves resource matching accuracy and scheduling efficiency by accurately understanding and quantifying the user's computing resource requirements through an agent, thereby generating a computing resource scheduling scheme.

Description

Technical Field

[0001] This invention belongs to the field of cloud computing resource management technology, and relates to a method and intelligent agent for identifying and recommending computing resource demand based on intelligent agents. In particular, it relates to a method and intelligent agent for human-computer interaction through artificial intelligence dialogue intelligent agents (AI Agents) to intelligently identify and evaluate user computing demand, and generate a computing resource scheduling recommendation scheme based on a multi-objective optimization algorithm. Background Technology

[0002] Currently, the mainstream methods for requesting and scheduling computing resources mainly include the following: (1) User self-selection: Users select from a complex list of resource specifications (such as vCPU cores, memory size, disk type, network bandwidth, etc.) through the cloud management platform interface. The platform may provide some basic price comparisons or simple recommendations based on historical usage records, but the decision depends entirely on the user's own professional knowledge.

[0003] (2) Work order consultation: Users submit work orders with incomplete descriptions. The operations or pre-sales engineers communicate with the users multiple times via email, instant messaging tools, etc., to repeatedly confirm details such as business scenarios, performance requirements, and budgets. The engineers then manually calculate resource requirements based on their experience and provide suggested solutions.

[0004] (3) Rule / template-based recommendation: The platform provides some preset resource configuration templates (such as "small application development environment", "high-concurrency web server", etc.), and users select according to the template name. This method lacks personalization and cannot adapt to complex and ever-changing business scenarios.

[0005] Based on the above three resource request and scheduling methods, the shortcomings of existing technologies are: (1) High threshold and low efficiency in resource application: Existing solutions require users to have a strong technical background, otherwise it is easy to select the wrong resource type or specification, resulting in resource waste or insufficient performance. Alternatively, it relies on manual intervention, which has high communication costs, long cycles, slow response speed, and inconsistent engineer experience levels, making it difficult to guarantee the quality of recommendations.

[0006] (2) One-sided understanding of requirements, easily overlooking key information: Existing solutions lack proactive and in-depth guidance mechanisms. Users (especially non-technical users) often cannot clearly and comprehensively describe their real needs (such as future concurrency, data processing logic, etc.), resulting in a mismatch between the resources requested and the actual business needs.

[0007] (3) Single recommendation dimension and lack of global optimization: Most existing recommendations only focus on whether the resource specifications are met or whether the price is the lowest, and lack a multi-dimensional comprehensive consideration from the perspective of global performance. For example, they fail to comprehensively consider the overall load balancing of the cluster, energy consumption (PUE), geographical distance between users and resources (network latency), and total cost of ownership (TCO) in the long term, and thus cannot achieve the optimal resource scheduling.

[0008] (4) Unfriendly interactive experience: Filling out complicated forms or having multiple rounds of human communication results in a poor user experience, which does not conform to the current technological development trend dominated by natural interaction.

[0009] In view of the shortcomings of the existing technologies, how to provide a computing resource scheduling method that can lower the threshold for users, improve the accuracy of resource matching and scheduling efficiency, and optimize the overall resource utilization and energy efficiency of the data center has become an urgent problem to be solved. Summary of the Invention

[0010] To address the problems in the background technology, this invention provides a method and agent-based approach for identifying and scheduling computing resource needs. An AI agent is constructed that integrates natural language processing (NLP), a knowledge base, a resource evaluation model, and a multi-objective optimization algorithm. Based on this agent, through natural and efficient human-computer dialogue, it accurately understands and quantifies the user's fuzzy computing resource needs. Furthermore, based on multi-dimensional and multi-objective optimization strategies, it automatically generates the optimal computing resource scheduling scheme for the user from a global perspective. This reduces the user's learning curve, improves resource matching accuracy and scheduling efficiency, and optimizes the overall resource utilization and energy efficiency of the data center.

[0011] To achieve the above objectives, this invention provides a method for identifying and scheduling computing resource requirements based on intelligent agents, comprising: Receive the user's fuzzy computing resource requirement description through a natural language interaction interface; Utilize natural language processing technology to analyze user needs and identify user intent and key entities; Based on the key entities and the built-in knowledge base, the system actively guides users to complete the missing parameters through multi-turn dialogues. The knowledge base includes typical resource requirement models and a list of key parameters for business scenarios. Once all parameters have been collected, the resource assessment model is invoked to quantify and calculate the parameters, generating a list of computing power resource requirements. Obtain a candidate resource set that matches the computing resource demand list from the computing power management platform; the candidate resource set includes heterogeneous computing power resources. For each candidate resource in the candidate resource set, a multi-dimensional utility score is calculated, wherein the multi-dimensional score includes at least two dimensions of resource load, cost, energy consumption, and network distance. Multiple recommendation schemes are generated based on different weighting strategies across multiple dimensions, and a natural language description of the recommendation reason is generated for each recommendation scheme. The recommended scheme is presented to the user, and resource allocation is performed based on the user's selection.

[0012] As a further improvement of the present invention, the key entity is compared with the knowledge base to identify the corresponding resource demand model and key parameters. If the identified parameters do not meet the requirements of the resource demand model, the agent initiates a multi-turn dialogue.

[0013] As a further improvement of the present invention, the multi-turn dialogue guidance steps are dynamically adjusted, specifically including: Identify missing parameters; Targeted questions are initiated based on missing parameters, and the subsequent question sequence is dynamically adjusted based on user responses to efficiently complete the parameters.

[0014] As a further improvement of the present invention, the resource assessment model is based on at least one of queuing theory, stress testing empirical formulas, or machine learning prediction models, and is used to output quantified CPU, memory, disk I / O, and network bandwidth requirements.

[0015] As a further improvement of the present invention, the calculation of the multi-dimensional utility score adopts a weighted function, and the weighting strategy includes at least two of the following: default comprehensive mode, cost priority mode or performance priority mode, and the weights can be configured based on user preferences.

[0016] As a further improvement of the present invention, the step of generating the recommendation scheme includes: Provide clear resource specifications, geographical location, price, and reasons for recommendation for each option; The reasons for the recommendations are generated based on a comparison of multi-dimensional utility scores to enhance the transparency of decision-making.

[0017] The present invention also provides an intelligent agent, comprising: an interaction module, an NLP processing module, a knowledge base module, a resource evaluation module, a resource scheduling optimization module, and a platform interface module; The interaction module is used for: Engage with users through a natural language interface; The NLP processing module is used for: Analyze user needs and identify intents and entities; The knowledge base module is used for: Storage business scenario resource requirement model and parameter list; The resource assessment module is used for: Quantify and calculate the complete parameters; The resource scheduling optimization module is used for: Perform multi-dimensional utility assessments and generate solutions; The platform interface module is used for: Interact with the computing power management platform to obtain candidate resources and execute allocations; The modules are connected via data flow, and the dialogue context is passed between the modules to drive the entire process.

[0018] As a further improvement of the present invention, the knowledge base module supports dynamic updates to adapt to new business scenarios; the resource scheduling optimization module integrates multi-objective optimization algorithms.

[0019] As a further improvement of the present invention, the inter-module data flow of the intelligent agent includes the transmission of dialogue context, parameter list and multi-dimensional utility score, and the data flow is orchestrated based on workflow engine to ensure state consistency.

[0020] As a further improvement of the present invention, the energy consumption dimension in the multi-dimensional utility evaluation is calculated based on the PUE value of the data center, and the network distance dimension prioritizes resources that are geographically close to the user to reduce latency. Compared with the prior art, the beneficial effects of the present invention are as follows: This invention fully automates the traditional process that relies on human expert experience, and achieves 24 / 7 instant response and service through intelligent agents. The high degree of intelligence and automation greatly improves efficiency and reduces labor costs.

[0021] This invention, through interactive guidance and knowledge base support, can deeply explore users' potential and real needs, accurately and comprehensively identify needs, avoid resource misallocation caused by information asymmetry, and improve the accuracy and rationality of resource applications.

[0022] This invention breaks through the limitations of making recommendations based on a single user or a single dimension (such as price). It takes a holistic view from the perspective of data center operators, taking into account multiple factors such as load, energy consumption, and cost. The recommendation strategy is globally optimal, thereby improving the overall efficiency of social resource utilization and achieving green energy-saving goals.

[0023] With this invention, users do not need to understand complex technical parameters. They can obtain professional and reliable personalized solutions simply through natural language dialogue and have decision-making power based on multi-angle information. The experience is more intuitive, user-friendly and efficient, and the user experience is significantly improved.

[0024] This invention provides clear reasons for each recommended solution, making the decision-making process transparent and credible, thus enhancing the reliability of the solutions and user trust, and solving the problem of unexplainable decision-making caused by the "black box" nature of AI models. Attached Figure Description

[0025] Figure 1This is a flowchart of a computing resource demand assessment method disclosed in one embodiment of the present invention; Figure 2 This is a flowchart of a resource scheduling scheme recommendation and allocation process disclosed in one embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1 , 2 As shown, the computing resource demand identification and scheduling recommendation method based on intelligent agents provided by this invention is based on the construction of an intelligent agent (AI Agent), which integrates natural language processing (NLP), a knowledge base, a resource evaluation model, and a multi-objective optimization algorithm. The overall process is divided into two main stages: the demand clarification and quantification stage and the resource matching and recommendation stage.

[0028] Phase 1: Demand Clarification and Quantification Phase, such as Figure 1 As shown, it includes: S1. Receive the user's fuzzy computing resource requirement description through a natural language interaction interface; Among them, the user interaction interface is deployed as a natural language interaction interface, where users make requests for computing resources to the intelligent agent through natural language; for example: I want to deploy an e-commerce website backend API service that can support 1,000 people online.

[0029] S2. Use natural language processing technology to analyze user needs and identify user intent and key entities; Among them, the intelligent agent uses NLP technology to parse user queries and identify user intent (such as "request resources") and key entities (such as "e-commerce website", "1000 people online", "API service").

[0030] S3. Based on key entities and a built-in knowledge base, it proactively guides users to complete missing parameters through multi-turn dialogues. The knowledge base includes typical resource requirement models and a list of key parameters for business scenarios. The process involves comparing key entities with a knowledge base to identify corresponding resource requirement models and key parameters. If the identified parameters do not meet the requirements of the resource requirement model, the agent initiates a multi-round dialogue.

[0031] Furthermore, the multi-round dialogue facilitation steps are dynamically adjusted, specifically including: Identify missing parameters; Targeted questions are initiated based on missing parameters, and the subsequent question sequence is dynamically adjusted based on user responses to efficiently complete the parameters.

[0032] Specifically, the intelligent agent compares the identified entities with a built-in knowledge base. This knowledge base contains typical resource requirement models and key parameter lists for various business scenarios (such as web applications, databases, and AI training).

[0033] If the identified parameters do not meet the requirements of the resource assessment model (e.g., missing "peak concurrent requests" or "average response time"), the agent will initiate a multi-turn dialogue, proactively and specifically asking questions to complete the information. For example, the agent might ask: "To calculate more accurately for you, what is your expected average number of requests per second (QPS)?" or "In your API service, does a single request typically require a complex database query? This helps in estimating CPU time." The entire process is guided, and the AI ​​dynamically adjusts subsequent questions based on the user's answers to ensure that all necessary parameters (such as quantitative indicators of CPU, memory, disk I / O, and network bandwidth) are collected efficiently.

[0034] S4. Once all parameters have been collected, the resource assessment model is invoked to quantify and calculate the parameters, generating a list of computing power resource requirements.

[0035] The resource assessment model is based on at least one of queuing theory, stress test empirical formulas, or machine learning prediction models, and is used to output quantified CPU, memory, disk I / O, and network bandwidth requirements.

[0036] Specifically, once all parameters have been collected, the agent invokes the built-in resource assessment model (e.g., based on queuing theory, stress test empirical formulas, or machine learning prediction models) to perform rigorous calculations and output a precise, quantified list of computing resource requirements (e.g., requiring a cloud server with 4 CPU cores, 8GB of memory, 50GB of SSD disk, and bandwidth ≥5Mbps). Phase Two: Resource Matching and Recommendation Phase, such as Figure 2 As shown, it includes: S5. Obtain a candidate resource set that meets the computing resource demand list from the computing power management platform. The candidate resource set includes heterogeneous computing power resources. The intelligent agent interacts with the underlying computing power management platform through the API interface, submits the above-mentioned quantified computing power resource demand list, and obtains a list of all available computing power resources that meet the conditions on the platform (candidate resource set), which may include heterogeneous resources (such as container instances, virtual machines, bare metal servers).

[0037] S6. Calculate a multi-dimensional utility score for each candidate resource in the candidate resource set. The multi-dimensional score includes at least two of the following dimensions: resource load, cost, energy consumption, and network distance. The calculation of multi-dimensional utility scores uses a weighted function, and the weighting strategy includes at least two of the following: default comprehensive mode, cost-first mode, or performance-first mode, and the weights can be configured based on user preferences.

[0038] Specifically, the agent performs a multi-dimensional utility evaluation of the candidate resource set. This process involves not simply listing candidate resources, but rather calculating a comprehensive utility score for each candidate resource. This score is a weighted function of multiple dimensions, primarily including: Resource load: Prioritize resources with lower load in the physical cluster to facilitate cluster load balancing.

[0039] Cost: Calculate the total cost based on on-demand / annual / monthly subscription prices.

[0040] Energy consumption: Based on the PUE (Power Usage Effectiveness) value of the data center where the resource is located, resources in data centers with higher energy efficiency are given priority.

[0041] Distance (network latency): Prioritize resources that are geographically closer to the target user to reduce network latency.

[0042] S7. Generate multiple recommendation schemes based on different weight strategies across multiple dimensions, and generate natural language descriptions of the recommendation reasons for each scheme; The agent generates multiple (at least two) different recommendation schemes based on different weight preferences (e.g., default comprehensive mode, cost-first mode, performance-first mode).

[0043] Furthermore, the steps for generating a recommendation include: Provide clear resource specifications, geographical location, price, and reasons for recommendation for each option; The recommendation rationale is generated based on a comparison of multi-dimensional utility scores to enhance decision-making transparency.

[0044] Specifically, each recommended option includes clear details (such as specific specifications, location, and price) and detailed reasons for the recommendation.

[0045] For example, Option 1 (lowest cost): "We recommend using Type A cloud servers located in the East China data center, with a monthly cost of XXX yuan. This option has the lowest cost while meeting performance requirements." Option 2 (Optimal Performance): "We recommend using a bare metal server (Type B) located in the South China data center (closest to your user base), with an estimated network latency of <10ms. Although the cost is slightly higher, it provides ultimate performance and stability, suitable for your core business." S8. Present recommended solutions to users and perform resource allocation based on user selections.

[0046] In this process, users select the solution that best suits their needs based on multi-faceted comparison information and recommendation reasons provided by the intelligent agent. After user confirmation, the intelligent agent can automatically call the API of the computing power management platform to complete the creation and allocation of resources.

[0047] The present invention also provides an intelligent agent, comprising: an interaction module, an NLP processing module, a knowledge base module, a resource evaluation module, a resource scheduling optimization module, and a platform interface module; Interactive module, used for: Engage with users through a natural language interface; NLP processing module, used for: Analyze user needs and identify intents and entities; The knowledge base module is used for: The system includes a resource requirement model and parameter list for storage business scenarios; the knowledge base module supports dynamic updates to adapt to new business scenarios; and the resource scheduling optimization module integrates multi-objective optimization algorithms.

[0048] The resource assessment module is used for: Quantify and calculate the complete parameters; The resource scheduling optimization module is used for: Perform multi-dimensional utility assessment and solution generation; among them, the energy consumption dimension in the multi-dimensional utility assessment is calculated based on the PUE value of the data center, and the network distance dimension prioritizes recommending resources that are geographically close to the user in order to reduce latency.

[0049] The platform interface module is used for: Interact with the computing power management platform to obtain candidate resources and execute allocations; The modules are connected via data flow, and the dialogue context is passed between the modules to drive the entire process. Furthermore, the data flow between the modules of the agent includes the transmission of dialogue context, parameter list and multi-dimensional utility score, and the data flow is orchestrated based on the workflow engine to ensure state consistency.

[0050] Advantages of this invention: This invention receives ambiguous requests through natural language interaction, identifies missing parameters using a knowledge base, and proactively initiates multi-round, dynamic, and targeted questioning to guide users in completing the information. It uses a pre-set resource evaluation model to calculate the collected complete parameters, generating a quantified request list. It then obtains a set of candidate resources that meet the quantified requirements from a computing platform, and comprehensively evaluates the utility of each candidate resource from multiple dimensions such as load, cost, energy consumption, and distance. Based on different weighting strategies, it generates multiple recommended solutions and provides natural language descriptions of the recommendation reasons for each solution. This overcomes the shortcomings of existing technologies, such as high resource application barriers, one-sided understanding of requests, single recommendation dimensions, and unfriendly interactive experiences. It achieves highly intelligent and automated processes, accurate and comprehensive request identification, globally optimal recommendation strategies, significantly improved user experience, and a transparent and reliable decision-making process. Thus, while lowering the user's barrier to entry, it improves resource matching accuracy, scheduling efficiency, and the overall performance of the data center.

[0051] The overall architecture of the intelligent agent of this invention includes an interaction module, an NLP processing module, a knowledge base module, a resource evaluation module, a resource scheduling optimization module, and an interface module with the computing power management platform. It integrates natural language processing (NLP), knowledge base, resource evaluation model and multi-objective optimization algorithm. Through the transmission of dialogue context between different modules and the driving of the process, it realizes the clarification and quantification of computing power resource requirements, resource matching and recommendation, and improves the accuracy of resource matching and scheduling efficiency.

[0052] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and scheduling computing resource requirements based on intelligent agents, characterized in that, include: Receive the user's fuzzy computing resource requirements through a natural language interaction interface; Utilize natural language processing technology to analyze user needs and identify user intent and key entities; Based on the key entities and the built-in knowledge base, the system actively guides users to complete the missing parameters through multi-turn dialogues. The knowledge base includes typical resource requirement models and a list of key parameters for business scenarios. Once all parameters have been collected, the resource assessment model is invoked to quantify and calculate the parameters, generating a list of computing power resource requirements. Obtain a candidate resource set that matches the computing resource demand list from the computing power management platform; the candidate resource set includes heterogeneous computing power resources. For each candidate resource in the candidate resource set, a multi-dimensional utility score is calculated, wherein the multi-dimensional score includes at least two dimensions of resource load, cost, energy consumption, and network distance. Multiple recommendation schemes are generated based on different weighting strategies across multiple dimensions, and a natural language description of the recommendation reason is generated for each recommendation scheme. The recommended scheme is presented to the user, and resource allocation is performed based on the user's selection.

2. The method for identifying and scheduling computing resource requirements based on intelligent agents according to claim 1, characterized in that: The key entities are compared with the knowledge base to identify the corresponding resource demand models and key parameters. If the identified parameters do not meet the requirements of the resource demand model, the agent initiates a multi-round dialogue.

3. The method for identifying and scheduling computing resource requirements based on intelligent agents according to claim 1, characterized in that, The multi-turn dialogue guidance steps are dynamically adjusted, and specifically include: Identify missing parameters; Targeted questions are initiated based on missing parameters, and the subsequent question sequence is dynamically adjusted based on user responses to efficiently complete the parameters.

4. The method for identifying and scheduling computing resource requirements based on intelligent agents according to claim 1, characterized in that, The resource assessment model is based on at least one of queuing theory, stress testing empirical formulas, or machine learning prediction models, and is used to output quantified CPU, memory, disk I / O, and network bandwidth requirements.

5. The method for identifying and scheduling computing resource requirements based on intelligent agents according to claim 1, characterized in that, The multi-dimensional utility score is calculated using a weighted function. The weighting strategy includes at least two of the following: default comprehensive mode, cost-first mode, or performance-first mode. The weights can be configured based on user preferences.

6. The method for identifying and scheduling computing resource requirements based on intelligent agents according to claim 1, characterized in that, The steps for generating the recommendation scheme include: Provide clear resource specifications, geographical location, price, and reasons for recommendation for each option; The reasons for the recommendations are generated based on a comparison of multi-dimensional utility scores to enhance the transparency of decision-making.

7. An intelligent agent for implementing the intelligent agent-based computing resource demand identification and scheduling recommendation method as described in any one of claims 1 to 6, characterized in that, include: The system includes an interaction module, an NLP processing module, a knowledge base module, a resource evaluation module, a resource scheduling optimization module, and a platform interface module. The interaction module is used for: Engage with users through a natural language interface; The NLP processing module is used for: Analyze user needs and identify intents and entities; The knowledge base module is used for: Storage business scenario resource requirement model and parameter list; The resource assessment module is used for: Quantify and calculate the complete parameters; The resource scheduling optimization module is used for: Perform multi-dimensional utility assessments and generate solutions; The platform interface module is used for: Interact with the computing power management platform to obtain candidate resources and execute allocations; The modules are connected via data flow, and the dialogue context is passed between the modules to drive the entire process.

8. The intelligent agent according to claim 7, characterized in that, The knowledge base module supports dynamic updates to adapt to new business scenarios; the resource scheduling optimization module integrates multi-objective optimization algorithms.

9. The intelligent agent according to claim 7, characterized in that, The inter-module data flow of the intelligent agent includes the transmission of dialogue context, parameter list and multi-dimensional utility score, and the data flow is orchestrated based on workflow engine to ensure state consistency.

10. The intelligent agent according to claim 7, characterized in that, The energy consumption dimension in the multi-dimensional utility assessment is calculated based on the PUE value of the data center, and the network distance dimension prioritizes resources that are geographically close to the user to reduce latency.

Images