Cloud computing resource automatic optimization scheduling system and method based on artificial intelligence

Abstract

The invention relates to the technical field of cloud computing, and discloses a cloud computing resource automatic optimization scheduling system and method based on artificial intelligence. According to the automatic optimal scheduling system and method for the cloud computing resources based on artificial intelligence, by intelligently selecting the migration opportunity of the virtual machine and the target host machine, the computing resources can be more reasonably distributed among different host machines, the situation that part of host resources are overloaded while part of host resources are idle is avoided, and the system and method are more efficient and more efficient. The utilization rate of overall computing resources is improved, it is ensured that each virtual machine can obtain relatively balanced resource support, the stability and performance of the system are improved, and the system can know possible traffic peaks in advance by accurately predicting real-time load data in the cloud environment, so that resource allocation and preparation are carried out in advance, and the service life of the system is prolonged. According to the invention, burst traffic can be effectively dealt with, service performance reduction and even system crash caused by sudden increase of traffic are avoided, continuity and stability of services are guaranteed, and user experience is improved.

Description

Technical Field

[0001] This application relates to the technical field of cloud computing, and particularly to a cloud computing resource automatic optimization scheduling system and method based on artificial intelligence. Background Art

[0002] A cloud computing resource automatic optimization scheduling system and method based on artificial intelligence aims to utilize advanced AI technology to intelligently manage resource allocation and task scheduling in a cloud computing platform. The system intelligently selects the virtual machine (VM) migration timing and target host in the cloud computing environment to solve the problem of uneven computing resources; at the same time, it adopts a precise prediction model for real-time load data to deal with the service performance degradation caused by sudden traffic, thereby ensuring service quality and user experience. In addition, it can automatically identify and model the characteristics of different types of application programs to ensure a high degree of matching between resource allocation and application requirements. To reduce energy consumption, the system has an effective mechanism for identifying and recycling idle computing resources, and through the strategy of dynamically adjusting task scheduling priorities, it can optimize the response time of critical services in a complex and changing load environment. Generally speaking, this system comprehensively considers multiple key challenges in cloud computing resource management and optimization scheduling and provides targeted solutions. Summary of the Invention

[0003] To solve the problems raised in the above background art, this application provides a cloud computing resource automatic optimization scheduling system and method based on artificial intelligence.

[0004] This application provides a cloud computing resource automatic optimization scheduling system and method based on artificial intelligence, adopting the following technical solutions:

[0005] A cloud computing resource automatic optimization scheduling method based on artificial intelligence includes:

[0006] S101. Collect and analyze the real-time load data of each host in the cloud environment, and use deep learning algorithms to predict future load trends;

[0007] S102. Identify the hosts with uneven computing resources according to the prediction results, and extract the running characteristics of different types of application programs;

[0008] S103. Build an application program feature model and match a suitable resource allocation plan to meet application requirements;

[0009] S104. Based on the resource allocation plan and load prediction, dynamically adjust the virtual machine migration timing, target host, and task scheduling priorities, and at the same time identify and recycle idle computing resources to improve platform energy efficiency.

[0010] Preferably, the future virtual machine migration timing and target host are intelligently determined based on the evaluation of the current and past migration effects of the virtual machine, and this evaluation depends on:

[0011] Compare the load balancing improvement and application response latency before and after the virtual machine migration, so as to calculate the benefit score R of each migration action;

[0012] Establish a historical data model to predict the impact of migrating to a specific host on the overall system efficiency;

[0013] Use machine learning algorithms to predict the optimal host allocation scheme, taking into account the expected load and network bandwidth limitations at different times;

[0014] Compare the effects of multiple candidate migration schemes, and adopt the formula max(C)>D = C*α + T*(1 - α), where C represents the benefit score, α is the weight factor representing the importance ratio of economic benefits, and T means the migration cost score.

[0015] Preferably, accurate cloud environment load prediction is carried out through deep reinforcement learning to improve the ability to cope with sudden traffic, including the following steps:

[0016] During the model training process, historical load peak information P and corresponding influencing factors such as weather changes Q and social hot events H are added;

[0017] Use a recurrent neural network architecture to capture the trend pattern M of long time intervals;

[0018] Introduce multi-source external variables to adjust the accuracy of the short-term fluctuation prediction value V;

[0019] If |Vpred - Vall| > τ, then output a warning signal and start additional emergency capacity deployment measures, where τ refers to the allowable error limit, and |...| represents the absolute value operation; Vpred represents the load prediction value, and Vall is all the recorded actual values.

[0020] Preferably, improve the resource configuration decision logic based on the application feature automatic recognition mechanism, and the specific implementation means are:

[0021] Collect the behavior logs B and configuration files S of various programs during operation to extract representative features X;

[0022] Apply clustering algorithms to map these features into different type labels T within a finite set;

[0023] Dynamically adapt to the most suitable computing environment E (such as GPU / CPU instances) according to the application type and user access mode;

[0024] When θX >= ρ(T), the corresponding rule chain will be activated to perform rapid resource adjustment operations to better meet the special task requirements. Here, θ is used to measure the matching strength under specific class labels, and ρ(T) defines a threshold value to ensure the selection of application category T that meets specific criteria.

[0025] Preferably, the implementation details of the efficient management and re-integration strategy for idle resources are as follows:

[0026] Regularly inventory the active process table L and resource occupancy details O within the cluster to discover potential recyclable objects Y;

[0027] Analyze long-term idle metrics such as CPU idle rate Icpu, memory utilization Imm, and disk idle ratio Idi sk to form a portrait of idle assets;

[0028] Select the best candidate Z that meets the conditions W (determined by characteristics such as the idle state maintenance period) from all the identified candidate devices to reduce operating costs;

[0029] Perform idle asset cleaning actions when the remaining ratio Rr of the resource pool is greater than the threshold ηR and the current task queue quantity Qt is less than the minimum tolerable waiting number ξQt, that is, (Rr > ηR) ∧ (Qt < ξQt) = true, then start the cleaning.

[0030] Preferably, to achieve more optimal dynamic management of task scheduling priorities to optimize the response speed of critical services, the measures include but are not limited to the following points:

[0031] Calculate the urgency index Ui of each task as the basis for preliminary screening;

[0032] Set the hierarchical quality of service SLA to weigh the importance Fi and feedback it to the task classifier J to guide the final sorting order Pi;

[0033] Sort the pending jobs according to the default probability Di in the service level agreement;

[0034] If Wi >= ψDi (Wi is the waiting cost coefficient, and ψ represents the sensitivity parameter reflecting the impact of delay losses), grant the emergency execution permission to quickly resolve the backlogged tasks and ensure the service quality level.

[0035] An artificial intelligence-based automatic optimization and scheduling system for cloud computing resources, the system is executed by the described artificial intelligence-based automatic optimization and scheduling method for cloud computing resources, and the system includes:

[0036] The analysis module is responsible for collecting a large amount of data from various components and links of the cloud computing system, including but not limited to hardware resource data such as the CPU usage rate, memory occupancy, network bandwidth, and storage I / O of the server, as well as business-related data such as the running status of virtual machines, the performance metrics of application programs, and the request frequency and types of users. These data come from a wide range of sources and have diverse formats. The analysis module can collect them uniformly, providing a comprehensive data basis for subsequent analysis work;

[0037] The identification module can accurately identify various hardware resources in the cloud computing environment, including the model of the server, configuration parameters (such as the number of CPU cores, memory size, and storage capacity), the specifications and performance metrics of network devices, and the type and read / write speed of storage devices. By identifying these hardware resources, the system can comprehensively understand the basic information of the underlying hardware, providing basic data support for subsequent resource scheduling and allocation. For example, based on the CPU performance of the server, it can determine which computing tasks are suitable to be allocated to that server;

[0038] The matching module, according to the identification results of the identification module on the task type and resource requirements, precisely matches the resource types and quantities required by the task with the available resources in the cloud computing system. For example, for compute-intensive tasks, the matching module will look for servers with strong CPU performance and idle computing resources; for memory-intensive tasks, it will focus on matching nodes with sufficient memory to ensure that the tasks can run in a suitable resource environment and improve the execution efficiency;

[0039] The dynamic adjustment module. In the cloud computing environment, resource requirements and system status change at any time. This module can dynamically adjust the resource allocation strategy based on the real-time data provided by the analysis module, such as resource utilization rate and task execution progress. For example, during an e-commerce promotion event when the volume of transaction processing tasks surges, it can timely increase computing and storage resources for related services to ensure the stable operation of the services.

[0040] In summary, this application includes at least one of the following beneficial technical effects:

[0041] 1. This cloud computing resource automatic optimization and scheduling system and method based on artificial intelligence can make the computing resources be more reasonably allocated among different host machines by intelligently selecting the virtual machine migration timing and target host machine, avoiding the situation where some host machines are overloaded with resources while some host machines have idle resources, improving the overall utilization rate of computing resources, ensuring that each virtual machine can obtain relatively balanced resource support, and enhancing the stability and performance of the system.

[0042] 2. The automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence can accurately predict the real-time load data in the cloud environment, enabling the system to anticipate possible traffic peaks in advance. Thereby, resource allocation and preparation can be carried out in advance, such as increasing server resources and adjusting network bandwidth in advance, effectively coping with sudden traffic, avoiding service performance degradation or even system crashes caused by sudden increases in traffic, ensuring the continuity and stability of services, and improving the user experience.

[0043] 3. The automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence can automatically identify and model the characteristics of different types of application programs, enabling the system to accurately allocate resources according to the actual needs of the application programs, avoiding waste caused by excessive resource allocation or affecting application performance due to insufficient allocation, improving the utilization efficiency of resources, enabling application programs to run in a suitable resource environment, enhancing the running efficiency and quality of applications, and giving full play to the advantages of the cloud computing platform.

[0044] 4. The automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence can effectively identify and recycle idle computing resources, timely shut down or adjust those unnecessary computing resources, avoiding their energy consumption in the idle state, thereby significantly reducing the energy consumption of the cloud computing platform, achieving energy conservation and emission reduction, reducing operating costs, and at the same time helping to improve the resource management efficiency of the cloud computing platform, enabling resources to be more concentratedly used for business with demands.

[0045] 5. The automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence optimize the dynamic adjustment strategy of task scheduling priorities, ensuring that in complex and changing load situations, critical services can always obtain a higher scheduling priority and receive the support of system resources first, thereby greatly shortening the response time of critical services, ensuring the efficient operation of critical services, improving the support ability of the entire system for critical services, and enhancing the reliability and availability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of an automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence according to the present invention.

[0047] Figure 2 It is a flowchart of an automatic optimization and scheduling system and method for cloud computing resources based on artificial intelligence according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following details the embodiments of the present application, and the examples of the embodiments are shown in the drawings.

[0049] In the description of this specification, the description with reference to the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0050] Next, refer to the attached Figure 1 , describing an artificial intelligence-based automatic optimization and scheduling system and method for cloud computing resources of the present invention. First, the system periodically collects various load parameters of each host machine and virtual machine in the cloud environment through distributed proxy nodes. The collected content includes but is not limited to data on CPU utilization, memory utilization, network bandwidth occupancy, disk read and write speed, etc. The collected real-time load data will be summarized and sent to a centralized management module, where a data analysis process will be started. Using deep learning algorithms such as LSTM recurrent neural networks or other time series prediction models, a time series analysis framework is established based on past historical load conditions. The framework aims to explore the regular characteristics within the load data and try to predict the changing trend of the task volume that each host machine may carry within a certain window period in the future (such as ten minutes to one hour).

[0051] In order to accurately solve the problem of service performance degradation caused by burst traffic, for example, in one embodiment, for a backend server group of an Internet financial platform, during the period from midnight to early morning when user access frequency is high, if a surge in the number of requests is detected in a short period of time, an appropriate amount of reserve virtual machines are deployed to the relevant host machines a few minutes in advance to divert new business requests and ensure that all customer order transactions can be completed smoothly. This approach ensures that even during peak periods, the service will not be stuck or crashed due to insufficient hardware resources, which not only improves the user experience but also protects the interests of merchants.

[0052] Next, based on the results of the prediction of future load trends, physical machines with resource redundancy or severe shortages are identified. These hosts may cause an imbalance in computing power within a local area, thereby affecting the overall working efficiency and service level of the cluster. It is very important to take targeted adjustment measures to address this situation. When the system determines that several nodes currently have excessive idle resources while other nodes are under heavy pressure, it is necessary to find appropriate solutions to balance them. For this purpose, application program operation status indicators such as the number of processes and the intensity of thread activities running in different virtual machines are extracted as objects for feature extraction. Then, with the help of supervised classification models or unsupervised pattern recognition techniques such as clustering algorithms, potential patterns are found from a large number of similar software operation records to form a specific application program feature representation. Through the above means, we can master the unique properties of each program to better plan their respective resident environments and execution paths.

[0053] After constructing the application program feature model, the next step is to determine the optimal resource allocation method. The main purpose of this step is to find the optimal solution that is most suitable for a specific workflow or task execution environment, thereby achieving higher-level service quality standard requirements. Specifically, the data set abstracted into mathematical formulas is mapped into a structured representation form that can be parsed and calculated according to predefined rules, and then through an optimization solver, an ideal configuration parameter combination scheme that conforms to the principle of maximizing economic benefits is selected. This can effectively prevent the phenomenon of low actual output value caused by blind allocation. For example, specifically, a certain big data mining team often uses a high-performance computer cluster to run complex machine learning training sets to generate iterative versions. At this time, the customized analysis tool they use will automatically detect the project scale and allocate the number of dedicated acceleration cards as needed to ensure a fast model convergence speed while avoiding the occurrence of computing power waste problems and ensuring that the resource utilization rate reaches the expected target value or above.

[0054] Furthermore, after mastering a reasonable application-level resource ratio plan, it is necessary to make necessary virtual machine migration strategy arrangements based on the conclusions drawn from the previous steps and in combination with the actual situation, that is, to determine when to migrate which virtual machine to which destination and appropriately adjust the priority sequence of new jobs waiting to be allocated to users. The key here is to grasp the timing and select the appropriate transfer point without affecting normal operation and without additionally increasing the overhead cost of the migration action itself. On the one hand, through in-depth analysis of the predicted information, make full preparations before the time period most likely to experience overloading. On the other hand, refer to the values obtained from real-time monitoring, flexibly respond to the uncontrollable factors brought about by changes in external conditions, maintain flexibility, and make reaction actions at any time to ensure system stability.

[0055] In addition, it is also necessary to actively monitor the status of idle units within the entire IT facility, proactively reclaim physical machines that have nothing to do, release the software and hardware associated identifiers they hold, and make them part of the candidate inventory list available for new demands, reducing power consumption and enhancing the overall benefit performance. For example, during the break period of an education cloud platform, some multimedia playback auxiliary teaching servers suspend most of their non-essential components to reduce standby power consumption, achieving energy conservation and emission reduction effects. At the same time, sufficient elastic space is reserved to quickly activate and restore all original functional states during the next peak class period without the need for large-scale temporary hardware expansion procurement, reducing a part of unnecessary expenses.

[0056] Finally, in the face of the increasingly complex coexistence situation of mixed business types, a dynamic adaptive priority adjustment mechanism must be formulated. It allows administrators to customize the relative importance and urgency level relationships between different types of work tasks and continuously perceive external pressure source stimuli through a feedback control system, timely updating decision results to ensure that core operational affairs are always on the optimal response path and not restricted by surrounding uncertain elements. For example, on an Internet of Things monitoring platform, for high-priority alarm events such as emergency help buttons from security monitoring cameras, the highest-level quick intervention power is given, which can immediately start querying the background database for video recording backtracking, linking with the 110 command center to dispatch police for support, and delaying the completion of ordinary inspection and measurement report generation and printing tasks to highlight the leading role of key business processes and demonstrate emergency response capabilities.

[0057] In one embodiment, a cloud computing resource automatic optimization and scheduling system based on artificial intelligence is also disclosed. This system is executed through the described cloud computing resource automatic optimization and scheduling method based on artificial intelligence, as Figure 2 shown, the system includes:

[0058] An analysis module, responsible for collecting a large amount of data from various components and links of the cloud computing system, including but not limited to hardware resource data such as the CPU usage rate, memory occupancy, network bandwidth, and storage I / O of servers, as well as business-related data such as the running status of virtual machines, performance indicators of application programs, and the request frequency and type of users. These data sources are extensive and in various formats, and the analysis module can uniformly collect them to provide a comprehensive data basis for subsequent analysis work;

[0059] The recognition module can accurately identify various hardware resources in the cloud computing environment, including the models of servers, configuration parameters (such as the number of CPU cores, memory size, storage capacity), specifications and performance indicators of network devices, and the types and read / write speeds of storage devices. By identifying these hardware resources, the system can comprehensively understand the basic information of the underlying hardware, providing basic data support for subsequent resource scheduling and allocation. For example, based on the CPU performance of the server, it can determine which computing tasks are suitable to be allocated to this server;

[0060] The matching module, according to the recognition results of the recognition module on the task type and resource requirements, precisely matches the resource types and quantities required by the task with the available resources in the cloud computing system. For example, for compute-intensive tasks, the matching module will look for servers with strong CPU performance and available computing resources; for memory-intensive tasks, it will focus on matching nodes with sufficient memory to ensure that the tasks can run in a suitable resource environment and improve execution efficiency;

[0061] The dynamic adjustment module. In the cloud computing environment, resource requirements and system status are constantly changing. This module can dynamically adjust the resource allocation strategy based on the real-time data provided by the analysis module, such as resource utilization rate, task execution progress, etc. For example, during an e-commerce promotion event, when the volume of transaction processing tasks surges, it can timely increase computing and storage resources for relevant services to ensure the stable operation of the services.

[0062] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limitations on the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence, characterized in that: include: S101, collect and analyze the real-time load data of each host in the cloud environment, and use deep learning algorithms to predict future load trends; S102, identifying a host machine with unbalanced computing resources according to the prediction result, and extracting operation characteristics of different types of application programs; S103, building an application feature model and matching an appropriate resource allocation solution to meet application requirements; S104: Based on the resource allocation scheme and load prediction, dynamically adjust the virtual machine migration timing, target host machine and task scheduling priority, and identify and recycle idle computing resources to improve platform energy efficiency.

2. According to the method of automatic optimization and scheduling of cloud computing resources based on artificial intelligence in claim 1, it is characterized in that: Intelligently determine the timing and target host for future virtual machine migration based on the evaluation of the current and past virtual machine migration effects. This evaluation depends on: Compare the load balancing improvements and application response delays before and after virtual machine migration to calculate the benefit score R of each migration action; Build historical data models to predict the impact of migration to a specific host on overall system efficiency; Use machine learning algorithms to predict the optimal host allocation plan, taking into account the expected load and network bandwidth constraints at different times; To compare the effects of multiple candidate migration plans, the formula max(C)>D=C*α+T*(1α) is used, where C represents the benefit score, α is the weight factor representing the importance ratio of economic benefits, and T means the migration cost score.

3. The method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence according to claim 2 is characterized in that: Accurate cloud environment load prediction through deep reinforcement learning improves the ability to cope with sudden traffic, including the following steps: During the model training process, historical load peak information P and corresponding influencing factors such as weather changes Q and social hot events H are added; Using a recurrent neural network architecture to capture trend patterns M over long time intervals; Introduce multi-source external variables to adjust the accuracy of short-term volatility estimate V; If |Vpred Vall|>τ, an early warning signal is output and additional emergency capacity deployment measures are initiated, where τ refers to the allowable error limit, |...| indicates the absolute value operation; Vpred represents the load prediction value, and Vall is all the actual values ​​recorded.

4. The method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence according to claim 3 is characterized in that: Improve resource allocation decision logic based on automatic identification mechanism of application features. The specific implementation means are: Collect the behavior logs B and configuration files S of various programs during operation to extract representative features X; Apply clustering algorithms to map these features into different types of labels T within a finite set; Dynamically adapt the most suitable computing environment E (e.g. GPU / CPU instance) according to application type and user access pattern; When θX>=ρ(T), the corresponding rule chain will be activated to perform fast resource adjustment operations to better meet the needs of special tasks. θ is used to measure the matching strength under a specific class label, and ρ(T) defines a threshold value to ensure that the application category T that meets the specific criteria is selected.

5. The method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence according to claim 4 is characterized in that: The details of the efficient management and reintegration strategy for idle resources are as follows: Regularly check the cluster-wide active process table L and resource usage details O to find potential objects to be recycled Y; Analyze long-term idle indicators such as CPU idle rate Icpu, memory utilization Imm, and disk idle ratio Idisk to form an idle asset profile; Select the best candidate Z that meets the condition W (determined by characteristics such as the idle state maintenance period) from all the identified candidate devices to reduce operating costs; When the remaining ratio Rr of the resource pool is greater than the threshold ηR and the current number of queued tasks Qt is less than the minimum tolerable waiting number ξQt, the idle asset cleanup action is performed, that is, (Rr>ηR)∧(Qt<ξQt)=true, then the cleanup is started.

6. The method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence according to claim 5 is characterized in that: Implement better dynamic management of task scheduling priorities to optimize the response speed of key businesses. Measures include but are not limited to the following: Calculate the urgency index Ui of each task as the basis for preliminary screening; Set the hierarchical service quality SLA to weigh the importance Fi and feed it back to the task classifier J to guide the final arrangement order Pi; Sort the pending jobs according to the probability of default Di in the service level agreement; If Wi>=ψDi (Wi is the waiting cost coefficient, ψ represents the sensitivity parameter reflecting the impact of delay loss), emergency execution authority is granted to quickly resolve backlog tasks and ensure service quality levels.

7. The cloud computing resource automatic optimization and scheduling system based on artificial intelligence according to claim 1 is characterized in that: The system adopts the method for automatic optimization and scheduling of cloud computing resources based on artificial intelligence according to any one of claims 1 to 6, and the system includes: The analysis module is responsible for collecting a large amount of data from various components and links of the cloud computing system, including but not limited to hardware resource data such as server CPU usage, memory usage, network bandwidth, storage I / O, as well as business-related data such as the operating status of virtual machines, application performance indicators, user request frequency and type, etc. These data come from a wide range of sources and in various formats. The analysis module can collect them uniformly and provide a comprehensive data foundation for subsequent analysis work. The identification module can accurately identify various hardware resources in the cloud computing environment, including server models, configuration parameters (such as the number of CPU cores, memory size, storage capacity), network device specifications and performance indicators, and storage device types and read and write speeds. By identifying these hardware resources, the system can fully understand the basic information of the underlying hardware and provide basic data support for subsequent resource scheduling and allocation, such as deciding which computing tasks are suitable for allocation to the server based on the server's CPU performance. The matching module accurately matches the resource type and quantity required by the task with the available resources in the cloud computing system based on the identification results of the task type and resource requirements by the identification module. For example, for computing-intensive tasks, the matching module will look for servers with strong CPU performance and idle computing resources; for memory-intensive tasks, it will focus on matching nodes with sufficient memory to ensure that the task can run in a suitable resource environment and improve execution efficiency. Dynamic adjustment module: In cloud computing environment, resource demand and system status are changing at any time. This module can dynamically adjust resource allocation strategy based on real-time data provided by the analysis module, such as resource utilization, task execution progress, etc. For example, during e-commerce promotion activities, the transaction processing tasks increase dramatically. It can promptly increase computing and storage resources for related businesses to ensure stable business operation.

Images