KVM (Keyboard Video Mouse) fault prediction method and system based on feedback mechanism, medium and equipment

Abstract

The invention provides a KVM (Keyboard Video Mouse) fault prediction method and system based on a feedback mechanism, a medium and equipment, and the system can dynamically adjust resource distribution in a virtualization environment through adaptive machine learning and a cross-domain resource scheduling mechanism, and optimize the use efficiency of resources. Through accurate fault prediction and rapid repair, in combination with a multi-level feedback mechanism and a deep learning model, the system can accurately predict potential faults, and through rapid response of an automatic repair mechanism, the downtime of the system is reduced. Through intelligent and automatic resource management and fault repair, the dependence on manual operation is remarkably reduced, the operation and maintenance cost is reduced, and the management efficiency of the virtualization platform is improved.

Description

Technical Field

[0001] This application belongs to the field of KVM fault prediction, and particularly relates to a KVM fault prediction method, system, medium and device based on a feedback mechanism. Background Art

[0002] Currently, traditional KVM resource scheduling often schedules resources based on static rules or simple algorithms. Static rules can lead to over-allocation or under-allocation of resources and it is difficult to adapt to the dynamic load requirements of the virtualization environment in real time. At the same time, existing systems mostly adopt predefined rules and history-based prediction models, lacking the ability of real-time detection and rapid response to faults. When the resource requirements of a virtual machine exceed the capabilities of a single host, traditional systems often do not have an effective mechanism for cross-host resource scheduling, resulting in resource waste or overload. Summary of the Invention

[0003] This application proposes a KVM fault prediction method, system, medium and terminal device based on a feedback mechanism, aiming to solve the problems in existing KVM virtualization resource management, and improve fault prediction, resource scheduling, repair automation and system stability.

[0004] The first aspect of this application provides a KVM fault detection method based on a feedback mechanism, and the method includes:

[0005] Collect the original resource data of virtual machines and hosts in real time, and preprocess the original resource data;

[0006] Obtain a resource scheduling strategy based on a pre-trained resource scheduling model, and determine the optimal resource scheduling strategy in combination with a preset reward mechanism;

[0007] Divide the host and virtual machine resources into no less than two resource pools according to a preset construction rule, perform dynamic management according to a preset trigger condition, and optimize the scheduling strategy according to the resource utilization feedback;

[0008] Train a preset fault prediction model in combination with the historical resource usage data of the host and virtual machine collected and the system load pattern;

[0009] Receive the real-time data generated by the resource scheduling model when executing the resource scheduling strategy for updating the fault prediction model;

[0010] Feed back the result of the fault prediction model to the resource scheduling model to optimize the resource scheduling strategy;

[0011] Perform fault repair according to the prediction result of the fault prediction model in combination with a preset priority;

[0012] Among them, the original data includes CPU usage rate, memory occupancy, disk I / O, network traffic, task queue length, response time, and historical load trend.

[0013] Through the adaptive machine learning and cross-domain resource scheduling mechanism, the system can dynamically adjust resource allocation in the virtualized environment and optimize the resource utilization efficiency; through accurate fault prediction and rapid repair, combined with the multi-level feedback mechanism and deep learning model, the system can accurately predict potential faults and quickly respond through the automatic repair mechanism, reducing the system downtime. Through intelligent and automated resource management and fault repair, the present invention significantly reduces the dependence on manual operations, reduces the operation and maintenance costs, and improves the management efficiency of the virtualization platform.

[0014] In a possible implementation method of the first aspect, the preprocessing of the original resource data is specifically as follows:

[0015] Calculate the moving average of the resource usage data of the original resource data at each time point;

[0016] Calculate the peak load and average load through built-in functions;

[0017] Obtain the CPU and memory quota information of the virtual machine, and calculate the quota usage ratio in combination with the actual usage;

[0018] Perform normalization processing on the above data to make it conform to the input of the resource scheduling model.

[0019] In a possible implementation method of the first aspect, the obtaining of the resource scheduling strategy based on the pre-trained resource scheduling model and determining the optimal resource scheduling action in combination with the preset reward mechanism is specifically as follows:

[0020] Input the normalized resource data into the pre-constructed resource scheduling model to obtain the resource scheduling strategy;

[0021] Give a positive reward when the resource scheduling strategy makes the resource utilization rate close to / within the optimal value range;

[0022] Give a negative reward when the resource scheduling strategy makes the resource utilization rate far from the optimal value range.

[0023] In a possible implementation method of the first aspect, the preset construction rules include at least one of division by resource type, division by physical host, division by virtual machine priority, and mixed division.

[0024] In a possible implementation method of the first aspect, the dynamic management according to the preset trigger conditions is specifically as follows:

[0025] When there is a real-time load change or a predicted resource demand change, dynamic resource management is performed according to a preset decision.

[0026] When the host reaches a preset threshold and local resources cannot meet the demand, cross-host resource management is performed.

[0027] In a possible implementation method of the first aspect, the real-time data generated by the receiving resource scheduling model during the execution of the resource scheduling policy is used to update the fault prediction model. Specifically:

[0028] The fault prediction model is trained based on the collected real-time resource usage data, historical resource usage data of virtual machines and hosts, and system load patterns.

[0029] The real-time data and scheduling operations obtained by the receiving resource scheduling model are input into the fault prediction model, and the auxiliary model is continuously updated and iterated.

[0030] In a possible implementation method of the first aspect, the result of the fault prediction model is fed back to the resource scheduling model to optimize the scheduling policy. Specifically:

[0031] The potential fault precursors and abnormal patterns of resource usage identified by the fault prediction model are fed back to the resource scheduling model.

[0032] The resource scheduling model adjusts the scheduling policy with reference to the received fault information.

[0033] The above solution collects the original resource data of virtual machines and hosts in real time, preprocesses it, and obtains the resource scheduling policy based on the pre-trained resource scheduling model, which enables the system to dynamically adjust resource allocation according to the actual resource usage. The use of intelligent and automated resource management and fault repair significantly reduces the dependence on manual operations. The system can perform a series of operations such as automatically collecting data, preprocessing, obtaining the scheduling policy, updating the fault prediction model, and performing fault repair without manual real-time monitoring and intervention.

[0034] The second aspect of this application provides a KVM fault prediction system based on a feedback mechanism. The system includes: a data collection module, a resource scheduling module, a fault prediction module, and a fault repair module. Among them,

[0035] The data collection module is used to collect the original resource data of virtual machines and hosts in real time and preprocess the original resource data.

[0036] The resource scheduling module obtains the resource scheduling policy based on the pre-trained resource scheduling model and determines the optimal resource scheduling policy in combination with a preset reward mechanism.

[0037] Divide the host machine and virtual machine resources into at least two resource pools according to preset construction rules, dynamically manage according to preset trigger conditions, and optimize the scheduling strategy based on resource utilization feedback;

[0038] The fault prediction module trains a preset fault prediction model by combining the historical resource usage data of the host machine and virtual machine and the system load pattern collected;

[0039] Receive the real-time data generated during the execution of the resource scheduling strategy by the resource scheduling model for updating the fault prediction model;

[0040] The fault repair module performs fault repair according to the prediction result of the fault prediction model in combination with the preset priority.

[0041] A third aspect of this application provides a storage medium, which stores computer-readable program code, and when the computer-readable program code is executed, it implements the steps of a KVM fault prediction method based on a feedback mechanism according to any one of the embodiments of this application.

[0042] A fourth aspect of this application provides a terminal device, the device includes: a terminal device, including a processor and a memory, the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a KVM fault prediction method based on a feedback mechanism according to any one of the embodiments of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of this application, the following will briefly introduce the drawings required for the implementation. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 is a flowchart of a KVM fault prediction method based on a feedback mechanism provided by an embodiment of this application;

[0045] Figure 2 is a structural diagram of a KVM fault prediction system based on a feedback mechanism provided by an embodiment of this application; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] In order to make the purpose, technical solutions and advantages of the present invention clearer, the following will further describe the present invention in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] Embodiment 1

[0048] Please refer to Figure 1 as shown, a KVM fault prediction method based on a feedback mechanism provided by an embodiment of the present application, including steps S101 - S104 specifically as follows:

[0049] Step S101: Real - time collect the original resource data of the virtual machine and the host machine, and pre - process the original resource data:

[0050] Further, the pre - processing of the original resource data is specifically as follows:

[0051] Calculate the moving average value of the resource usage data of the original resource data at each time point;

[0052] Calculate the peak load and the average load through built - in functions;

[0053] Obtain the CPU and memory quota information of the virtual machine, and calculate the quota usage ratio in combination with the actual usage amount;

[0054] Perform standardization processing on the above - mentioned data so that it conforms to the input of the resource scheduling model.

[0055] Specifically, use the moving average method to calculate the long - term trend of resource usage. For the resource usage data within a period of time (such as the past 24 hours), calculate the average moving data at each time point. The specific calculation formula of the average moving data is:

[0056]

[0057] where n is the time window size, and x i is the historical resource usage data. In a specific embodiment, for the CPU usage rate data, set a window size (such as 12 time intervals), and calculate the average value of the CPU usage rate in the previous 12 time intervals at each time point.

[0058] Analyze the load fluctuation situation by calculating the standard deviation of the data. The standard deviation reflects the degree of dispersion of the data relative to the average value. The larger the standard deviation, the greater the load fluctuation; obtain the CPU and memory quota information of the virtual machine, calculate the quota usage situation of the CPU and memory, and obtain the usage ratio by dividing the actual usage amount by the quota amount. Perform standardization processing on the relevant data. Standardization processing can ensure that each eigenvalue is within a unified range, avoiding the influence of different dimensions of resource types on the subsequent model processing.

[0059] Step S102: Obtain the resource scheduling policy based on the pre - trained resource scheduling model, and determine the optimal resource scheduling policy in combination with the preset reward mechanism;

[0060] Divide the host and virtual machine resources into at least two resource pools according to the preset construction rules, perform dynamic management according to the preset trigger conditions, and optimize the scheduling strategy based on the resource utilization feedback;

[0061] Furthermore, the resource scheduling model based on pre-training obtains a resource scheduling strategy and determines the optimal resource scheduling action in combination with a preset reward mechanism. Specifically:

[0062] Input the standardized resource data into the pre-constructed resource scheduling model to obtain a resource scheduling strategy;

[0063] Give a positive reward when the resource scheduling strategy makes the resource utilization rate close to / within the optimal value range;

[0064] Give a negative reward when the resource scheduling strategy makes the resource utilization rate far from the optimal value range.

[0065] Exemplarily, input the standardized data into the pre-constructed resource scheduling model to obtain the corresponding resource scheduling strategy; when the resource scheduling strategy makes the resource utilization rate increase to reach the optimal value range, give a positive reward. For example, when the CPU and memory utilization rates of the virtual machine are close but do not exceed the reasonable range, the reward is set as: r1 = k1×(cpu utization +mem utilization ). Give a negative reward when the resource utilization rate is far from the optimal value range. For example, when the CPU utilization rate of the host exceeds 90%, the reward can be set as r2 = -k2×(cpu utization -0.9); give a positive reward when the resource scheduling strategy makes the load between hosts more balanced. Measure the load balance degree by calculating the load variance between hosts. The smaller the load variance, the higher the reward; give a negative reward when the load imbalance intensifies. Give a positive reward when the resource scheduling strategy ensures the service quality of the virtual machine, such as the network latency is within the acceptable range and the disk I / O response time is normal. When the network latency Latency is less than the preset threshold T, the reward can be set as r4 = k4×(T - Latency); resource migration (such as virtual machine migration) will bring certain costs, such as migration time, network bandwidth consumption, etc. Give a negative reward when resource migration occurs.

[0066] Furthermore, the preset construction rules include at least one of the following: division by resource type, division by physical host, division by virtual machine priority, and hybrid division.

[0067] It should be noted that when classifying by resource type, all homogeneous resources of all hosts and virtual machines are incorporated into the corresponding resource pools, including the CPU resource pool, memory resource pool, disk I / O resource pool, network bandwidth resource pool, etc. The total number of CPU cores of all hosts constitutes the global CPU resource pool. When classifying by powerless hosts, for example, the local CPU resource pool of host A and the local memory resource pool of host B. Each host maintains its own resource pool to preferentially ensure the needs of local virtual machines, and the remaining resources are added to the global resource pool. When classifying by virtual machine priority, independent resource pools are allocated according to the service level agreement (SLA) of the virtual machines to ensure the isolation of critical business resources. The hybrid classification can adopt a global CPU resource pool (by type) + regional memory resource pool (by physical location), taking into account both resource type and geographical location to reduce the cross-regional migration cost.

[0068] Furthermore, the dynamic management is carried out according to the preset trigger conditions, specifically:

[0069] When there is a real-time load change or a resource demand change is predicted, dynamic resource management is performed according to the preset decision.

[0070] When a host reaches the preset threshold and its local resources cannot meet the demand, cross-host resource management is carried out.

[0071] It should be noted that the resource management in load scheduling mainly targets the resource allocation within a single host (such as CPU and memory quota adjustment) or the resource optimization of a single virtual machine (such as enabling standby instances). When there is a real-time load change (such as the CPU usage rate of a virtual machine fluctuates) or a resource demand change is predicted, the resource quota is dynamically adjusted or a new instance is started by combining the current load and historical patterns. The virtual machine configuration can be directly modified or a new instance can be started through the virtualization platform API.

[0072] Specifically, cross-host resource management is triggered by resource thresholds. For example, the CPU utilization rate of a host ≥ 90% or the remaining memory ≤ 10%, the length of the virtual machine task queue continuously exceeds the threshold (such as the queue length > 100) or the response time > 500 ms and other hosts have idle resources (such as the CPU idle rate > 30% and the memory idle rate > 40%). Among them, cross-host resource management includes: real-time migration of virtual machines and dynamic invocation between resource pools. The target host can be selected through a cross-domain scheduling algorithm to perform data migration and resource reallocation.

[0073] The greedy algorithm based on the resource pool status preferentially selects a host machine with sufficient idle resources and low network latency, and uses the live migration technology of KVM (such as migration based on shared storage or pre-copy migration). During the migration process, the virtual machine service is kept uninterrupted, and at the same time, the migration cost (such as time and bandwidth consumption) is recorded. Preferably, the resource scheduling strategy is optimized by combining reinforcement learning. After a successful migration, if the resource utilization rate increases, a positive reward is set, and if the migration cost increases, a negative reward is set. The resource scheduling model is trained through historical migration data to learn the optimal migration trigger conditions and the target host selection strategy.

[0074] In a specific embodiment, taking the CPU resource pool as an example, the resource pool utilization threshold is set to 80%. When the resource pool utilization rate is greater than or equal to 80% and the duration exceeds 30 seconds, the corresponding resource scheduling strategy is triggered, such as local resource optimization, borrowing across resource pools, cross-host scheduling, etc. Local resource optimization can release CPU resources by shutting down unused virtual machine instances and preferentially compressing the idle resources in the local resources; when the CPU resources are insufficient, hyper-threading technology can be borrowed from the memory resource pool across resource pools, and resources can be borrowed from other types of resource pools; cross-host scheduling can migrate the virtual machine to a host machine with a CPU idle rate > 40%, and schedule from the global resource pool or the idle resource pool of other host machines. For example: the CPU resource pool utilization rate of host machine A reaches 90% and lasts for 45 seconds; check whether there is reserved CPU resource locally. If not, evaluate whether the memory resource pool can be borrowed and calculate the borrowing amount: demand gap = 90% - 80% = 10%, borrowing amount = 10% × conversion coefficient (such as 0.5) = 5% CPU resource; enable the memory over-allocation technology to temporarily increase the CPU resource; when the CPU utilization rate drops to 85%, the reward value r = +3, which is fed back to the model to record this strategy.

[0075] S103: Train a preset fault prediction model by combining the historical resource usage data of the host machine and the virtual machine and the system load pattern collected;

[0076] Receive the real-time data generated by the resource scheduling model during the execution of the resource scheduling strategy for updating the fault prediction model;

[0077] Based on the result of the fault prediction model, feedback to the resource scheduling model to optimize the resource scheduling strategy;

[0078] Furthermore, the receiving the real-time data generated by the resource scheduling model during the execution of the resource scheduling strategy for updating the fault prediction model is specifically:

[0079] Train the fault prediction model based on the collected real-time resource usage data, historical resource usage data of the virtual machine and the host machine, and the system load pattern;

[0080] The real-time data obtained by the resource scheduling model and the scheduling operation are input into the fault prediction model, assisting the model to continuously update and iterate.

[0081] It should be noted that a large amount of real-time data is generated during the process of the resource scheduling model performing resource allocation and scheduling operations, such as the migration records of virtual machines, the dynamic adjustment of resources, etc. These data can be fed back to the fault prediction model to help the fault prediction model continuously update and optimize, improving the accuracy of fault prediction. The result of resource scheduling will change the resource usage status and load distribution of the system, thereby affecting the input and prediction results of the fault prediction model. For example, a successful virtual machine migration operation may relieve the resource pressure of a certain host and reduce the probability of the host failing. The fault prediction model re-evaluates the possibility of a fault occurring based on the new resource status.

[0082] Furthermore, the result of the fault prediction model is fed back to the resource scheduling model to optimize the scheduling strategy. Specifically:

[0083] The potential fault precursors and abnormal patterns of resource usage identified by the fault prediction model are fed back to the resource scheduling model;

[0084] The resource scheduling model adjusts the scheduling strategy with reference to the received fault information.

[0085] It should be noted that when analyzing historical resource usage data, the fault prediction model can identify some potential fault precursors and abnormal patterns of resource usage. These information can be shared with the resource scheduling module and model. When making resource allocation and scheduling decisions, the resource scheduling model will refer to these potential fault information to avoid over-allocation of resources to virtual machines or hosts that may fail, and thus take preventive measures in advance, such as migrating tasks to more stable nodes. When the fault prediction model detects that a certain virtual machine or host has a potential fault, the resource scheduling module will adjust the scheduling strategy according to the feedback. For example, if it is predicted that the CPU resources of a certain host are about to have an overload fault, the resource scheduling module will migrate some virtual machines to other hosts with lower load in advance to avoid the occurrence of the fault.

[0086] The fault prediction model can evaluate the effect of resource scheduling. If after the scheduling operation, the fault prediction model detects that the fault risk of the system has decreased, it means that the scheduling strategy is effective. On the contrary, if the fault risk is still high or even increases, the scheduling strategy needs to be optimized and adjusted; the resource scheduling model continuously adjusts and optimizes the scheduling strategy according to the output result of the fault prediction model. For example, if the fault prediction model finds that a certain scheduling mode is likely to cause some resources to fail, the resource scheduling model will try to adopt other scheduling algorithms or rules to reduce the fault risk of the system.

[0087] The process of scheduling the load will change the resource usage status and load distribution of the system, generating more data samples of different types. These new data can enrich the training data of the fault prediction model, enabling the fault prediction model to learn more resource usage patterns and fault precursors, thereby improving the accuracy of fault prediction. Reasonable load scheduling can evenly distribute the resources of the system, avoid overuse of certain resources, and thus reduce the probability of faults occurring in the chest pain. For example, by migrating high-load tasks to other idle nodes, the resource pressure on a certain node can be relieved, reducing the faults caused by resource overload.

[0088] S104: Perform fault repair according to the prediction results of the fault prediction model in combination with the preset priority.

[0089] Specifically, when a resource bottleneck is detected or the fault prediction model gives a warning, the system will automatically execute repair strategies, such as readjusting resources, virtual machine migration, etc., to quickly restore the stability of the KVM system. Record the detailed logs of each repair operation in the system log file, and evaluate the repair effect through a real-time feedback mechanism. These data can be used as a reference for subsequent operations and provide input for the learning and optimization of the fault prediction model.

[0090] Preferably, set the corresponding priority rules, where the fault repair priority is higher than the resource scheduling priority. For example, when a hardware fault is detected and repair is triggered, suspend the regular resource scheduling; the repair operation may trigger temporary resource scheduling (such as migrating the faulty virtual machine), which needs to cooperate with the regular scheduling strategy.

[0091] Implementing the embodiments of the present application has the following effects:

[0092] Through the adaptive machine learning and cross-domain resource scheduling mechanism, the KVM system can dynamically adjust resource allocation in the virtualization environment, optimize the resource usage efficiency, and avoid overload or resource waste. Combine the multi-level feedback mechanism to accurately predict potential faults and quickly respond through the automatic repair mechanism, reducing the system downtime. The intelligent decision-making engine of the system dynamically adjusts resource and task allocation according to real-time feedback to ensure that the resources in the virtualization environment are always in the optimal configuration. Through intelligent and automated resource management and fault repair, this system significantly reduces the dependence on manual operations, reduces the operation and maintenance costs, and improves the management efficiency of the virtualization platform.

[0093] Embodiment 2

[0094] Please refer to Figure 2 , a KVM fault prediction system based on a feedback mechanism provided by the embodiments of the present application, including: a data acquisition module, a resource scheduling module, a fault prediction module, and a fault repair module, where

[0095] The data acquisition module is used to collect the original resource data of the virtual machine and the host in real time and preprocess the original resource data;

[0096] The resource scheduling module obtains a resource scheduling policy based on a pre-trained resource scheduling model, and determines an optimal resource scheduling policy in combination with a preset reward mechanism;

[0097] The host and virtual machine resources are divided into at least two resource pools according to a preset construction rule, dynamically managed according to a preset trigger condition, and the scheduling policy is optimized according to the resource utilization feedback;

[0098] The fault prediction module trains a preset fault prediction model in combination with the historical resource usage data of the host and the virtual machine collected and the system load pattern;

[0099] Receives the real-time data generated by the resource scheduling model during the execution of the resource scheduling policy, and is used to update the fault prediction model;

[0100] The fault repair module repairs faults according to the prediction results of the fault prediction model in combination with a preset priority.

[0101] The above KVM fault detection system based on a feedback mechanism can implement a KVM fault prediction method based on a feedback mechanism in the above method embodiment. The optional items in the above method embodiment are also applicable to this embodiment and will not be elaborated here. The remaining content of the embodiments of the present application can refer to the content of the above method embodiment. In some preferred embodiments, it will not be repeated.

[0102] Embodiment III

[0103] Correspondingly, the present application further provides a computer-readable storage medium, and the computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute a KVM fault prediction method based on a feedback mechanism as described in any one of the above embodiments.

[0104] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0105] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0106] The so-called processor may be a Central Processing Unit (CPU), or 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. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device through various interfaces and lines.

[0107] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by invoking the data stored in the memory, the processor realizes various functions of the terminal device. The memory may mainly include a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function, etc.; the data storage area can store the data created according to the use of the mobile terminal, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, memory, plug-in hard disks, Smart Media Cards (SMCs), Secure Digital (SD) cards, Flash Cards, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0108] Among them, if the modules / units integrated in the terminal device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of this application, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0109] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A KVM fault prediction method based on a feedback mechanism, characterized in that, Including: Real-time collect the original resource data of virtual machines and host machines, and preprocess the original resource data; Obtain a resource scheduling strategy based on a pre-trained resource scheduling model, and determine the optimal resource scheduling strategy in combination with a preset reward mechanism; Divide the host machine and virtual machine resources into at least two resource pools according to preset construction rules, perform dynamic management according to preset trigger conditions, and optimize the scheduling strategy according to the resource utilization feedback; Train a preset fault prediction model in combination with the historical resource usage data of the host machine and virtual machine and the system load pattern collected; Receive the real-time data generated by the resource scheduling model during the execution of the resource scheduling strategy for updating the fault prediction model; Feedback the result of the fault prediction model to the resource scheduling model to optimize the resource scheduling strategy; Perform fault repair according to the prediction result of the fault prediction model in combination with the preset priority; Wherein, the original data includes CPU usage rate, memory occupancy, disk I / O, network traffic, task queue length, response time, and historical load trend.

2. The KVM fault prediction method based on a feedback mechanism according to claim 1, wherein The preprocessing of the original resource data is specifically: Calculate the moving average of the resource usage data of the original resource data at each time point; Calculate the peak load and average load through built-in functions; Obtain the CPU and memory quota information of the virtual machine, and calculate the quota usage ratio in combination with the actual usage; Perform normalization processing on the above data to make it conform to the input of the resource scheduling model.

3. The KVM fault prediction method based on a feedback mechanism according to claim 1, wherein The obtaining of the resource scheduling strategy based on the pre-trained resource scheduling model and determining the optimal resource scheduling action in combination with the preset reward mechanism is specifically: Input the normalized resource data into a pre-constructed resource scheduling model to obtain a resource scheduling strategy; Give a positive reward when the resource scheduling strategy makes the resource utilization rate close to / within the optimal value range; Give a negative reward when the resource scheduling strategy makes the resource utilization rate far from the optimal value range.

4. The KVM fault prediction method based on a feedback mechanism according to claim 1, wherein The preset construction rules include at least one of: division by resource type, division by physical host machine, division by virtual machine priority, and mixed division.

5. The KVM fault prediction method based on a feedback mechanism according to claim 4, characterized in that, The dynamic management according to the preset trigger conditions is specifically: When there is a real-time load change or a predicted resource demand change, perform dynamic resource management according to the preset decision; When the host machine reaches the preset threshold and the local resources cannot meet the demand, perform cross-host machine resource management.

6. The KVM fault prediction method based on a feedback mechanism according to claim 1, wherein, The receiving of the real-time data generated by the resource scheduling model during the execution of the resource scheduling strategy for updating the fault prediction model is specifically: Train the fault prediction model based on the collected real-time resource usage data, historical resource usage data of virtual machines and host machines, and the system load pattern; Receive the real-time data and scheduling operations obtained by the resource scheduling model and input them into the fault prediction model to assist the model in continuous updating and iteration.

7. The KVM fault prediction method based on a feedback mechanism according to claim 6, wherein The feedback of the result of the fault prediction model to the resource scheduling model to optimize the scheduling strategy is specifically: Feedback the potential fault precursors and abnormal patterns of resource usage identified by the fault prediction model to the resource scheduling model; The resource scheduling model adjusts the scheduling strategy with reference to the received fault information.

8. A KVM fault prediction system based on a multi-level feedback mechanism, characterized in that, Including: A data collection module, a resource scheduling module, a fault prediction module, and a fault repair module, wherein, The data acquisition module is used to collect the original resource data of the virtual machine and the host in real time and preprocess the original resource data; The resource scheduling module obtains a resource scheduling policy based on a pre-trained resource scheduling model and determines an optimal resource scheduling policy in combination with a preset reward mechanism; The host and virtual machine resources are divided into at least two resource pools according to a preset construction rule, dynamically managed according to a preset trigger condition, and the scheduling policy is optimized according to the resource utilization feedback; The fault prediction module trains a preset fault prediction model in combination with the historical resource usage data of the host and virtual machine and the system load pattern collected; Receive the real-time data generated by the resource scheduling model during the execution of the resource scheduling policy for updating the fault prediction model; The fault repair module repairs faults according to the prediction results of the fault prediction model in combination with a preset priority.

9. A storage medium, characterized in that, The storage medium stores computer-readable program code, and when the computer-readable program code is executed, the steps of a KVM fault prediction method based on a feedback mechanism according to any one of claims 1 to 7 are implemented.

10. A terminal device, characterized in that, It includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, the steps of a KVM fault prediction method based on a feedback mechanism according to any one of claims 1 to 7 are implemented.

Images