Intelligent cloud desktop resource scheduling method and system with fusion of fault recovery

Abstract

The application relates to the technical field of cloud computing resource management, and discloses an intelligent cloud desktop resource scheduling method and system combined with fault recovery, which comprises the following steps: dynamically analyzing the resource supply and demand state of a cloud desktop cluster to obtain resource supply and demand characteristics and calculating the resource real-time scheduling pressure corresponding to the resource supply and demand characteristics; detecting the health degree of key nodes of the cloud desktop cluster to obtain node redundancy capability data; analyzing the resource recovery time efficiency of the cloud desktop cluster when a fault occurs, generating a fault recovery guarantee coefficient corresponding to the cloud desktop cluster, and dividing fault emergency grades corresponding to the fault recovery guarantee coefficient; configuring a multi-stage fault recovery plan corresponding to the cloud desktop cluster, locating a resource bottleneck area of the cloud desktop cluster when a fault occurs, calculating the resource diffusion efficiency corresponding to the resource bottleneck area, and formulating a resource optimization scheduling strategy combined with fault recovery corresponding to the cloud desktop cluster. The application aims to improve the scheduling accuracy of intelligent cloud desktop resources.

Description

Technical Field

[0001] This invention relates to an intelligent cloud desktop resource scheduling method and system for integrated fault recovery, belonging to the field of cloud computing resource management technology. Background Technology

[0002] Intelligent cloud desktop clusters are systems that use virtualization technology to centrally manage computing, storage, and other resources and allocate them to user terminals on demand. They are widely used in scenarios such as enterprise office, education and training, and remote collaboration. The rationality of their resource scheduling and their fault recovery capabilities directly affect user experience and business continuity.

[0003] Currently, there are two main methods for intelligent cloud desktop resource scheduling: one is static allocation, which allocates fixed resources to user terminals according to preset rules. This method has poor flexibility, low resource utilization, and cannot cope with dynamically changing resource demands. The other is scheduling based on simple load balancing, which adjusts resources by monitoring node load. However, it is insufficient in predicting potential fault risks, has slow recovery speed after a fault occurs, and is difficult to achieve accurate resource scheduling and efficient reuse. It cannot effectively solve problems such as system lag and business interruption caused by improper resource scheduling. Therefore, an intelligent cloud desktop resource scheduling method that integrates fault recovery is needed to ensure efficient scheduling of cloud desktop systems. Summary of the Invention

[0004] This invention provides a method and system for scheduling intelligent cloud desktop resources that integrates fault recovery, with the main purpose of improving the scheduling accuracy of intelligent cloud desktop resources.

[0005] To achieve the above objectives, the present invention provides an intelligent cloud desktop resource scheduling method for integrated fault recovery, comprising:

[0006] The system acquires real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, it dynamically analyzes the resource supply and demand status of the cloud desktop cluster to obtain resource supply and demand characteristics and calculates the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics.

[0007] Based on the real-time scheduling pressure of the resources, the potential fault risk points of the cloud desktop cluster are analyzed, the fault impact factors corresponding to the potential fault risk points are queried, and based on the fault impact factors, the health status of the key nodes of the cloud desktop cluster is detected to obtain node redundancy capability data.

[0008] Based on the node redundancy capability data, the resource recovery time of the cloud desktop cluster when a failure occurs is analyzed. Based on the resource recovery time and the current resource distribution status, a fault recovery guarantee coefficient corresponding to the cloud desktop cluster is generated, and the fault emergency level corresponding to the fault recovery guarantee coefficient is divided.

[0009] Based on the fault emergency level, configure a multi-level fault recovery plan corresponding to the cloud desktop cluster. Based on the multi-level fault recovery plan, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault and calculate the resource diffusion efficiency corresponding to the resource bottleneck area.

[0010] Based on the resource diffusion efficiency, a resource dynamic adjustment instruction corresponding to the cloud desktop cluster is generated, and the resource dynamic adjustment instruction is sent to the preset cloud platform management and control center to obtain the center execution feedback data. Based on the center execution feedback data, a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster is formulated.

[0011] Optionally, the step of dynamically analyzing the resource supply and demand status of the cloud desktop cluster based on the real-time load data and resource request information to obtain resource supply and demand characteristics includes:

[0012] Analyze the current resource supply capacity corresponding to the real-time load data;

[0013] Calculate the total user resource demand corresponding to the resource request information;

[0014] Analyze the distribution of demand types corresponding to the total user resource demand;

[0015] Based on the distribution of the demand types, identify the resource gap data corresponding to the cloud desktop cluster;

[0016] Based on the resource gap data, the resource supply and demand status of the cloud desktop cluster is dynamically analyzed to obtain resource supply and demand characteristics.

[0017] Optionally, calculating the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics includes:

[0018] The resource supply and demand characteristics are decomposed into dimensions to obtain a multidimensional supply and demand sequence;

[0019] Calculate the rate of change of demand and the rate of change of supply for each dimension in the multidimensional supply and demand sequence at adjacent time points;

[0020] Query the node resources corresponding to each dimension in the multidimensional supply and demand sequence, and collect the node hardware parameters and real-time load of the node resources.

[0021] By combining the node hardware parameters and the node's real-time load, the node's real-time response efficiency corresponding to the node resources is calculated.

[0022] Combining the demand change rate, the supply change rate, and the real-time response efficiency of the nodes, the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics is calculated using the following formula:

[0023]

[0024] Where A represents the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics. This represents the node resource weight corresponding to the d-th dimension in the multidimensional supply and demand sequence. This represents the rate of change of demand in the d-th dimension of a multidimensional supply and demand sequence at adjacent time points. This represents the rate of change in supply for the d-th dimension in a multidimensional supply and demand sequence at adjacent time points. This represents the baseline response efficiency corresponding to node resource e. Let t represent the real-time response efficiency of node resource e corresponding to the d-th dimension at time t, where ta and tb represent the supply and demand start time and supply and demand end time of the multidimensional supply and demand sequence, respectively, d represents the sequence number of the multidimensional supply and demand sequence, q represents the number of multidimensional supply and demand sequences, e represents the sequence number of the node resource, and r represents the number of node resources.

[0025] Optionally, the step of performing health checks on key nodes of the cloud desktop cluster based on the fault impact factors to obtain node redundancy capability data includes:

[0026] Query the health status detection standards corresponding to the aforementioned fault influencing factors;

[0027] Based on the health detection criteria, key nodes of the cloud desktop cluster are selected.

[0028] Based on the key nodes, monitor the health indicators corresponding to each dimension of the key nodes.

[0029] Calculate the deviation between the health index and the preset health benchmark value to obtain the node health deviation.

[0030] Based on the node health deviation, the health of key nodes in the cloud desktop cluster is checked to obtain node redundancy capability data.

[0031] Optionally, analyzing the resource recovery time of the cloud desktop cluster in the event of a failure based on the node redundancy capability data includes:

[0032] Extract the redundancy configuration features from the node redundancy capability data;

[0033] Based on the redundant configuration characteristics, the redundant coverage area corresponding to the cloud desktop cluster is determined;

[0034] Quantify the resource replenishment rate corresponding to the redundant coverage area;

[0035] Based on the resource replenishment rate, the local recovery differences corresponding to the cloud desktop cluster are evaluated;

[0036] Based on the aforementioned local recovery differences, the resource recovery time of the cloud desktop cluster during a failure is analyzed.

[0037] Optionally, the step of classifying the fault recovery guarantee coefficient into corresponding fault emergency levels includes:

[0038] Based on the fault recovery guarantee coefficient, query the key recovery features corresponding to the cloud desktop cluster;

[0039] Analyze the characteristic fluctuation index in the key recovery features;

[0040] Based on the characteristic fluctuation index, the core recovery range corresponding to the cloud desktop cluster is determined;

[0041] Generate the emergency response threshold corresponding to the core recovery range;

[0042] Based on the emergency response threshold, the emergency response level corresponding to the fault recovery guarantee coefficient is classified.

[0043] Optionally, locating the resource bottleneck area of ​​the cloud desktop cluster during a failure, based on the multi-stage fault recovery plan, includes:

[0044] Analyze the resource call paths corresponding to the multi-stage fault recovery plan;

[0045] Based on the resource call path, mark the key resource nodes of the cloud desktop cluster;

[0046] Filter the core monitoring dimensions of the key resource nodes and collect real-time fault data of the core monitoring dimensions;

[0047] Analyze the abnormal fluctuations in the real-time fault data;

[0048] Based on the abnormal fluctuations caused by the fault, the resource bottleneck area of ​​the cloud desktop cluster during the fault can be located.

[0049] Optionally, calculating the resource diffusion efficiency corresponding to the resource bottleneck region includes:

[0050] Obtain the regional resources corresponding to the resource bottleneck area, monitor and process the regional resources, and record the monitoring cycle of the regional resources;

[0051] Detect the availability of resources in the region during the monitoring period, and analyze the regional services corresponding to the resource bottleneck region;

[0052] Based on the regional services, calculate the resource priority corresponding to the regional resources;

[0053] Combining the resource priority, the monitoring period, and the resource availability, the resource diffusion efficiency corresponding to the resource bottleneck area is calculated using the following formula:

[0054]

[0055] Where F represents the resource diffusion efficiency corresponding to the resource bottleneck area. This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This represents the theoretical transfer rate of resources in the i-th region. This represents the resource priority corresponding to the i-th region resource, where i represents the sequence number of the region resource, and n represents the number of region resources. and This indicates the time point in the monitoring cycle.

[0056] Optionally, generating the resource dynamic adjustment instruction corresponding to the cloud desktop cluster based on the resource diffusion efficiency includes:

[0057] Analyze the resource flow range corresponding to the resource diffusion efficiency;

[0058] Based on the resource flow range, the resource sensitivity level corresponding to the cloud desktop cluster is divided;

[0059] Query the resource balancing data associated with changes in the resource sensitivity level;

[0060] Extract the efficiency regulation factors from the resource balance data;

[0061] Based on the performance control factor, a dynamic resource adjustment instruction corresponding to the cloud desktop cluster is generated.

[0062] To address the aforementioned problems, this invention also provides an intelligent cloud desktop resource scheduling system integrating fault recovery, the system comprising:

[0063] The scheduling pressure calculation module is used to obtain real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, it dynamically analyzes the resource supply and demand status of the cloud desktop cluster, obtains resource supply and demand characteristics, and calculates the real-time scheduling pressure of resources corresponding to the resource supply and demand characteristics.

[0064] The health detection module is used to analyze the potential fault risk points of the cloud desktop cluster based on the real-time scheduling pressure of the resources, query the fault impact factors corresponding to the potential fault risk points, and perform health detection on the key nodes of the cloud desktop cluster based on the fault impact factors to obtain node redundancy capability data.

[0065] The fault emergency response level classification module is used to analyze the resource recovery time of the cloud desktop cluster when a fault occurs based on the node redundancy capability data, generate the fault recovery guarantee coefficient corresponding to the cloud desktop cluster based on the resource recovery time and the current resource distribution status, and classify the fault emergency response level corresponding to the fault recovery guarantee coefficient.

[0066] The diffusion efficiency calculation module is used to configure a multi-level fault recovery plan corresponding to the cloud desktop cluster based on the fault emergency level, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault based on the multi-level fault recovery plan, and calculate the resource diffusion efficiency corresponding to the resource bottleneck area.

[0067] The scheduling strategy formulation module is used to generate a dynamic resource adjustment instruction corresponding to the cloud desktop cluster based on the resource diffusion efficiency, send the dynamic resource adjustment instruction to a preset cloud platform management and control center, obtain center execution feedback data, and formulate a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster based on the center execution feedback data.

[0068] Compared to the problems described in the background technology, this invention, by acquiring real-time load data of the intelligent cloud desktop cluster and resource request information from user terminals, can provide real-time and comprehensive basic data for subsequent resource scheduling analysis. This allows for a precise understanding of cluster resource usage and actual user needs, enabling more accurate resource supply and demand analysis and scheduling pressure calculation. Consequently, it improves the rationality of resource scheduling and the timeliness of fault recovery. Based on the real-time resource scheduling pressure, this invention analyzes the potential fault risk points of the cloud desktop cluster, accurately locating weak links that may fail under high scheduling pressure. It clarifies the risk differences of different nodes and resource types, providing crucial evidence for evaluating cluster stability and fault tolerance. Furthermore, based on the node redundancy capability data, this invention analyzes the resource recovery timeliness of the cloud desktop cluster during fault occurrences, allowing for in-depth analysis of node redundancy. This invention addresses the specific impact of point redundancy configuration on resource recovery speed after a fault, quantifies the recovery delay caused by insufficient redundancy capabilities, and further, configures multi-level fault recovery plans for the cloud desktop cluster based on the fault emergency level. This allows for the construction of differentiated recovery strategy systems for fault scenarios of varying urgency, from rapid self-healing processes for low-level faults to end-to-end redundancy switching for high-level faults, achieving precise fault response grading and improving the resilience and efficiency of the cloud desktop cluster in handling complex faults. Finally, based on the resource diffusion efficiency, this invention generates dynamic resource adjustment instructions for the cloud desktop cluster, accurately quantifying the resource flow efficiency in resource bottleneck areas and transforming complex diffusion data into intuitive resource allocation instructions. This helps identify resource scheduling bottleneck trends in advance, ensuring the stability and efficiency of cloud desktop cluster resource supply from the source. Therefore, the intelligent cloud desktop resource scheduling method and system with integrated fault recovery provided by this invention can improve the scheduling accuracy of intelligent cloud desktop resources. Attached Figure Description

[0069] Figure 1 A flowchart illustrating an intelligent cloud desktop resource scheduling method for integrated fault recovery provided in an embodiment of the present invention;

[0070] Figure 2 A flowchart illustrating the classification of fault emergency response levels is provided in one embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram of a module for implementing the intelligent cloud desktop resource scheduling system for fusion fault recovery, provided in an embodiment of the present invention.

[0072] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0073] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0074] This application provides an intelligent cloud desktop resource scheduling method with integrated fault recovery. The executing entity of the intelligent cloud desktop resource scheduling method with integrated fault recovery includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the intelligent cloud desktop resource scheduling method with integrated fault recovery can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0075] Example 1:

[0076] Reference Figure 1 The diagram shown is a flowchart illustrating an intelligent cloud desktop resource scheduling method for integrated fault recovery provided in an embodiment of the present invention. In this embodiment, the intelligent cloud desktop resource scheduling method for integrated fault recovery includes:

[0077] S1. Obtain real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, dynamically analyze the resource supply and demand status of the cloud desktop cluster to obtain resource supply and demand characteristics, and calculate the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics.

[0078] This invention, by acquiring real-time load data of the intelligent cloud desktop cluster and resource request information from user terminals, can provide real-time and comprehensive basic data for subsequent resource scheduling analysis, accurately grasp the usage status of cluster resources and the actual needs of users, and more accurately carry out resource supply and demand status analysis, scheduling pressure calculation, and other tasks, thereby improving the rationality of resource scheduling and the timeliness of fault recovery.

[0079] The intelligent cloud desktop cluster refers to a cloud desktop service cluster formed by integrating multiple physical servers through virtualization technology. It is used to provide computing, storage, and network resource support for a large number of user terminals. For example, a cloud desktop cluster deployed by an enterprise, containing 50 physical servers, can provide office desktop services for 2,000 employee terminals simultaneously. The real-time load data refers to real-time data reflecting the current resource usage of each node in the cloud desktop cluster, including CPU utilization, memory usage, disk I / O rate, network bandwidth usage, etc. For example, a node has a CPU utilization of 75%, a memory usage of 60%, and a network bandwidth usage of 40Mbps. The resource request information of the user terminal refers to various resource request commands and related parameters issued by the user during the use of the cloud desktop. For example, when a user opens a large design software, the user issues a request for increased CPU computing power, a request for increased memory, and information such as the amount of resources requested and the duration. Optionally, obtaining real-time load data of the intelligent cloud desktop cluster can be achieved through a distributed monitoring system, such as using Prometheus combined with Grafana to build a monitoring platform and collect resource usage data of each node in real time; obtaining resource request information from user terminals can be achieved through API interface listening technology, such as using the Java Spring Cloud framework to listen to the resource request APIs sent by user terminals and parse the request information including request type and resource quantity.

[0080] Based on the real-time load data and resource request information, this invention dynamically analyzes the resource supply and demand status of the cloud desktop cluster to obtain resource supply and demand characteristics. It can accurately reveal the matching rules and trends between the cluster's resource supply and user demand in real-time scenarios, providing a core basis for quantifying real-time resource scheduling pressure and assessing potential fault risk points. This enables resource scheduling to deeply reflect the dynamic process of cluster resource supply and demand, and improves the adaptability of scheduling strategies to actual operating conditions.

[0081] The resource supply and demand characteristics refer to the various changes and patterns in the supply and demand of resources in a cloud desktop cluster under the influence of real-time load and user requests. It is a general summary of the resource supply and demand relationship. For example, the cluster's CPU resource demand surges and supply becomes tight between 9-11 am on weekdays, while memory resource supply and demand are balanced between 3-5 pm. This information reflecting the dynamic process of resource supply and demand constitutes the resource supply and demand characteristics.

[0082] As an embodiment of the present invention, the step of dynamically analyzing the resource supply and demand status of the cloud desktop cluster based on the real-time load data and resource request information to obtain resource supply and demand characteristics includes:

[0083] Analyze the current resource supply capacity corresponding to the real-time load data;

[0084] Calculate the total user resource demand corresponding to the resource request information;

[0085] Analyze the distribution of demand types corresponding to the total user resource demand;

[0086] Based on the distribution of the demand types, identify the resource gap data corresponding to the cloud desktop cluster;

[0087] Based on the resource gap data, the resource supply and demand status of the cloud desktop cluster is dynamically analyzed to obtain resource supply and demand characteristics.

[0088] The current resource supply capacity refers to the total amount and distribution of various resources available to each node of the cloud desktop cluster at the time of analysis. It represents the current state of resource supply. For example, the cluster currently provides 200GHz of idle CPU computing power, 500GB of idle memory, and 10TB of idle disk space. These data constitute the current resource supply capacity. The total user resource demand refers to the sum of resource requests issued by all user terminals within a certain period of time, reflecting the overall scale of user demand for resources. For example, the total CPU resource demand of user terminals in a certain hour is 300GHz, and the total memory demand is 600GB. B; The demand type distribution refers to the proportion of different types of resources in the total demand of users, reflecting the demand intensity of various resources. For example, CPU resource demand accounts for 40%, memory resource demand accounts for 30%, network bandwidth demand accounts for 20%, disk I / O demand accounts for 10%, etc.; The resource gap data refers to the part of the cloud desktop cluster's current resource supply capacity that cannot meet the total demand of users, reflecting the imbalance between resource supply and demand. For example, the CPU resource gap is 100GHz, the memory resource gap is 100GB, etc. Recording the gap values ​​and differences of these different types of resources forms the resource gap data.

[0089] Optionally, the analysis of the current resource supply capacity corresponding to the real-time load data can be achieved through resource aggregation algorithms, such as using the MapReduce framework to aggregate and calculate the load data of each node, and finally obtaining the total current resource supply capacity of the cluster; the statistics of the total user resource demand corresponding to the resource request information can be achieved through data aggregation technology, such as using SQL statements to aggregate and count the user request records stored in the database, and finally obtaining the total user resource demand; the analysis of the demand type distribution corresponding to the total user resource demand can be achieved through classification and statistical technology, such as using Python's NumPy library to classify and count different types of resource demands, and finally obtaining the demand type distribution; the identification of the resource gap data corresponding to the cloud desktop cluster can be achieved through difference calculation technology, such as calculating the difference between the total user resource demand and the current resource supply capacity item by item, and finally obtaining the resource gap data; the dynamic analysis of the resource supply and demand status of the cloud desktop cluster can be achieved through time series analysis technology, such as using the ARIMA model to perform time series modeling of resource supply and demand data, and finally obtaining resource supply and demand characteristics containing time-varying features.

[0090] This invention quantifies the scheduling tension of cloud desktop clusters under the influence of resource supply and demand by calculating the real-time scheduling pressure corresponding to the resource supply and demand characteristics. It provides key indicators for assessing the rationality of resource scheduling and the risk of failure. This pressure can intuitively reflect the urgency of resource scheduling and help analyze the potential failure risk level.

[0091] The real-time resource scheduling pressure refers to the average intensity / pressure of resource allocation by the cloud desktop cluster to meet user resource requests per unit time under dynamic resource supply and demand scenarios. It quantifies the tension in the resource scheduling process by integrating multi-dimensional data such as real-time load changes, user request frequency, and resource type differences. It is a key indicator reflecting the cluster's resource scheduling capability. The larger the value, the greater the scheduling pressure, and the more likely the cluster is to experience resource allocation imbalance or failure.

[0092] As an embodiment of the present invention, the calculation of the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics includes:

[0093] The resource supply and demand characteristics are decomposed into dimensions to obtain a multidimensional supply and demand sequence;

[0094] Calculate the rate of change of demand and the rate of change of supply for each dimension in the multidimensional supply and demand sequence at adjacent time points;

[0095] Query the node resources corresponding to each dimension in the multidimensional supply and demand sequence, and collect the node hardware parameters and real-time load of the node resources.

[0096] By combining the node hardware parameters and the node's real-time load, the node's real-time response efficiency corresponding to the node resources is calculated.

[0097] Combining the demand change rate, the supply change rate, and the real-time response efficiency of the nodes, the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics is calculated using the following formula:

[0098]

[0099] Where A represents the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics. This represents the node resource weight corresponding to the d-th dimension in the multidimensional supply and demand sequence. This represents the rate of change of demand in the d-th dimension of a multidimensional supply and demand sequence at adjacent time points. This represents the rate of change in supply for the d-th dimension in a multidimensional supply and demand sequence at adjacent time points. This represents the baseline response efficiency corresponding to node resource e. Let t represent the real-time response efficiency of node resource e corresponding to the d-th dimension at time t, where ta and tb represent the supply and demand start time and supply and demand end time of the multidimensional supply and demand sequence, respectively, d represents the sequence number of the multidimensional supply and demand sequence, q represents the number of multidimensional supply and demand sequences, e represents the sequence number of the node resource, and r represents the number of node resources.

[0100] The multidimensional supply and demand sequence is a structured data sequence obtained by decomposing resource supply and demand characteristics according to resource type (e.g., CPU, memory) and time window. The demand change rate and supply change rate are the ratios of demand difference to time interval and supply difference to time interval for each dimension in the multidimensional supply and demand sequence, respectively. The node resources are the specific amount of resources that can be allocated to each node in the cluster for each dimension in the multidimensional supply and demand sequence. The node hardware parameters and node real-time load are the node's own performance indicators (e.g., CPU frequency, memory capacity) and current resource usage (e.g., CPU utilization, memory occupancy) corresponding to the node resources, respectively. The node real-time response efficiency is the ability value of the node resources to respond to user requests per unit time, calculated by combining the node hardware parameters and the node real-time load. The baseline response efficiency is the theoretical maximum response efficiency of the node under ideal conditions (e.g., no load).

[0101] Optionally, the resource supply and demand characteristics can be dimensionally decomposed using a dual-dimensional segmentation technique based on resource type and time window (e.g., segmentation by resource type such as CPU and memory, and a 10-minute time window) to obtain a multidimensional supply and demand sequence. The demand change rate and supply change rate for each dimension in the multidimensional supply and demand sequence at adjacent times can be calculated using a time difference algorithm (e.g., calculating the ratio of the demand difference between adjacent 5-minute intervals to the time interval). The node resources corresponding to each dimension in the multidimensional supply and demand sequence can be queried using a cluster resource mapping table (e.g., a database table containing the correspondence between each resource dimension and nodes). Hardware detection tools and negative... The monitoring plugin (such as using ipmitool to detect hardware parameters, or the Zabbix plugin to monitor load) collects the node hardware parameters and real-time load corresponding to the node resources; combining the node hardware parameters and the node real-time load, the real-time response efficiency corresponding to the node resources is calculated using the performance calculation formula (such as node real-time response efficiency = theoretical hardware performance × (1 - real-time load rate)); the node resource weight is obtained by comprehensively evaluating the node's hardware configuration (such as the number of CPU cores and memory size), historical load stability, and support for key businesses, using a weighted calculation model (such as the analytic hierarchy process).

[0102] The steps for real-time resource scheduling pressure calculation are as follows: Decompose the resource allocation across six dimensions, including CPU, memory, and network bandwidth, to obtain a multi-dimensional supply and demand sequence (q=6); collect the demand change rate over adjacent 5-minute intervals. and rate of change in supply For example, in the CPU dimension, ΔB = 15% / min and ΔD = 8% / min; node hardware parameters include a CPU clock speed of 3.2GHz and a memory bandwidth of 20GB / s. Real-time load data is collected every second through Prometheus to calculate the node's real-time response efficiency. When substituting into the formula, The CPU dimension weight is set to 0.3, βe (baseline response efficiency) is taken as 80% of the theoretical hardware value, the integration interval [ta,tb] is set to 10 minutes, and the scheduling pressure value A is finally obtained. When A>0.7, a fault risk warning is triggered.

[0103] It should be noted that the formula for calculating real-time resource scheduling pressure treats the resource scheduling pressure of the cloud desktop cluster as the result of the interaction between supply and demand changes in different resource dimensions and node response capabilities. It calculates the difference in supply and demand change rates (demand change rate minus supply change rate) for each resource dimension (such as CPU, memory, etc.) using node resource weights as coefficients, then divides this difference by the weighted sum of the node resource's baseline response efficiency and real-time response efficiency. Finally, it integrates over the supply and demand time interval to achieve a mathematical model of dynamic scheduling pressure. Its core assumption is that the scheduling pressure of each resource dimension can be linearly superimposed, and that node response efficiency reflects the actual capacity for resource allocation.

[0104] In practical applications, such as enterprise office cloud desktop cluster scheduling, this formula can decompose complex resource supply and demand scenarios into scheduling contributions from various resource dimensions. For example, in scenarios with both office software (consuming CPU) and big data analysis (consuming memory) tasks, it accurately distinguishes between CPU scheduling pressure and memory scheduling pressure. Testing shows that for the CPU resource dimension, the scheduling pressure calculation error has decreased from ±20% to ±8% (actual application verification is needed). By adjusting node resource weights and response efficiency parameters, it adapts to the resource requirements of different businesses. For scenarios with resource demand fluctuations of up to 50%, the pressure decoupling accuracy is improved by 25%, effectively identifying potential failure risks caused by resource bottlenecks and providing more accurate data for cloud desktop cluster fault recovery and resource optimization scheduling.

[0105] S2. Based on the real-time scheduling pressure of the resources, analyze the potential fault risk points of the cloud desktop cluster, query the fault impact factors corresponding to the potential fault risk points, and perform health checks on the key nodes of the cloud desktop cluster based on the fault impact factors to obtain node redundancy capability data.

[0106] Based on the real-time scheduling pressure of the resources, this invention analyzes the potential failure risk points of the cloud desktop cluster, accurately locates the weak links that may fail under high scheduling pressure, clarifies the risk differences of different nodes and resource types, and provides key basis for evaluating the stability and fault tolerance of the cluster.

[0107] The potential fault risk points refer to the specific locations or resource links where the cloud desktop cluster may fail under real-time resource scheduling pressure. These include node hardware anomalies, resource allocation imbalances, and network link congestion, reflecting its vulnerability under high load scheduling. For example, if a node's CPU utilization is consistently above 90%, it may crash when scheduling pressure surges. This node is a potential fault risk point. Optionally, the analysis of the potential fault risk points of the cloud desktop cluster can be achieved through fault tree analysis technology, such as using FTA software combined with Python's NetworkX library to construct a fault propagation model, ultimately obtaining potential fault risk points including node overload and network congestion.

[0108] This invention can transform qualitative risk situations into quantitative indicators by querying the fault impact factors corresponding to the potential fault risk points, providing a unified reference for evaluating the impact of faults on cluster services. It can accurately associate risk points with consequences such as service interruption and data loss, helping to quickly determine their priority in scenarios such as fault recovery and resource redundancy configuration.

[0109] The fault impact factor refers to an indicator used to quantify the impact of potential fault risk points on cluster operation. It is derived by a specific algorithm based on factors such as the location of the risk point, the scope of impact, and business relevance, and reflects the severity of the fault consequences. For example, by analyzing the business interruption duration and the number of affected users caused by a core node failure, a value of 7.5 is calculated using a formula. This 7.5 is the fault impact factor, reflecting the degree of harm caused by the failure. Optionally, the query for the fault impact factor corresponding to the potential fault risk point can be achieved through fault mode and impact analysis techniques, such as using FMEA tables combined with the analytic hierarchy process (AHP) to calculate the risk priority number, and finally obtaining the fault impact factor that characterizes the degree of fault impact.

[0110] Based on the aforementioned fault impact factors, this invention performs health checks on key nodes of the cloud desktop cluster to obtain node redundancy capability data. It can accurately assess the health status and redundancy reserves of key nodes for different impact levels, revealing the replacement capability and recovery potential of nodes when a fault occurs. This allows the detection to move from risk point identification to node redundancy dynamics, improving the pertinence and effectiveness of fault recovery preparation.

[0111] The node redundancy capability data refers to a set of data that reflects the redundancy support capability and recovery guarantee of key nodes in the cloud desktop cluster during failure, which is formed by integrating information such as the health score of each key node, the capacity of backup resources, and the fault switching speed. For example, a redundancy capability assessment report containing characteristics such as a core node health score of 90, backup node resources that can support 80% of services, and a switching time of 2 seconds is node redundancy capability data.

[0112] As an embodiment of the present invention, the step of performing health checks on key nodes of the cloud desktop cluster based on the fault impact factor to obtain node redundancy capability data includes:

[0113] Query the health status detection standards corresponding to the aforementioned fault influencing factors;

[0114] Based on the health detection criteria, key nodes of the cloud desktop cluster are selected.

[0115] Based on the key nodes, monitor the health indicators corresponding to each dimension of the key nodes.

[0116] Calculate the deviation between the health index and the preset health benchmark value to obtain the node health deviation.

[0117] Based on the node health deviation, the health of key nodes in the cloud desktop cluster is checked to obtain node redundancy capability data.

[0118] The health detection standard refers to the threshold used to divide detection items and judge health status based on the critical node health assessment criteria pre-set according to the fault impact factor. For example, when the fault impact factor is 8, the critical node CPU health threshold is set to utilization rate ≤70% and the memory health threshold is utilization rate ≤60%. If the threshold is exceeded, it is judged as a health abnormality. The critical node refers to the core node that is crucial to the operation of business in the cloud desktop cluster according to the health detection standard. It usually includes servers that carry critical business, network hub nodes, etc. For example, database server nodes and core switch connection nodes are selected as critical nodes in the enterprise cloud desktop cluster, and their health is tested in a targeted manner. The health index refers to the health data actually measured by detection equipment at various dimensions of key nodes, reflecting the current real operating status of the nodes. For example, if a server monitoring tool measures a key node's CPU utilization of 65%, memory usage of 55%, and network latency of 10ms, this is the node's health index. The node health deviation refers to the data formed by comparing the health index of the key node with a preset health benchmark value one by one, and arranging the calculated deviation in dimensional order. This data is used to characterize the degree to which the node's health status deviates from the normal level. For example, if the benchmark value is CPU utilization ≤60% and memory usage ≤50%, and the measured deviation is +5% and +5%, then the node health deviation data is formed.

[0119] Optionally, querying the health detection standards corresponding to the fault influencing factors can be achieved through expert system knowledge base technology, such as using the CLIPS expert system combined with an industry standard library to match detection thresholds, ultimately obtaining detection standards that distinguish different health levels; screening the key nodes of the cloud desktop cluster can be achieved through business dependency graph analysis technology, such as using the Neo4j graph database to construct a node-business association graph, ultimately obtaining key nodes based on business importance; monitoring the health indicators corresponding to each dimension of the key nodes can be achieved through multi-dimensional monitoring technology, such as using the Zabbix monitoring system combined with the Prometheus indicator collector, ultimately obtaining accurate health indicators for each dimension; calculating the health indicators and preset health benchmark values ​​can be achieved through deviation analysis algorithms, such as calculating the relative deviation based on the standard deviation standardization method on the R language platform, ultimately obtaining deviation data reflecting the differences in node health; health detection of the key nodes of the cloud desktop cluster can be achieved through health assessment model technology, such as integrating hardware status, load data, and response speed and processing them through a BP neural network algorithm, ultimately obtaining node redundancy capability data including redundancy capabilities.

[0120] S3. Based on the node redundancy capability data, analyze the resource recovery time of the cloud desktop cluster when a failure occurs. Based on the resource recovery time and the current resource distribution status, generate the fault recovery guarantee coefficient corresponding to the cloud desktop cluster, and classify the fault emergency level corresponding to the fault recovery guarantee coefficient.

[0121] Based on the node redundancy capability data, this invention analyzes the resource recovery time of the cloud desktop cluster when a failure occurs, and can deeply analyze the specific impact of node redundancy configuration on the resource recovery speed after a failure, and quantify the degree of recovery delay caused by insufficient redundancy capability.

[0122] The resource recovery time refers to the time required for a cloud desktop cluster to restore normal resource supply after a failure, taking into account the impact of node redundancy capabilities. It is usually expressed as the time from the occurrence of the failure to the full recovery of resources (such as minutes or hours). For example, if a cluster experiences a node failure and restores all resource supply within 30 minutes with the help of a backup node, its resource recovery time is 30 minutes.

[0123] As an embodiment of the present invention, the step of analyzing the resource recovery time of the cloud desktop cluster in the event of a failure based on the node redundancy capability data includes:

[0124] Extract the redundancy configuration features from the node redundancy capability data;

[0125] Based on the redundant configuration characteristics, the redundant coverage area corresponding to the cloud desktop cluster is determined;

[0126] Quantify the resource replenishment rate corresponding to the redundant coverage area;

[0127] Based on the resource replenishment rate, the local recovery differences corresponding to the cloud desktop cluster are evaluated;

[0128] Based on the aforementioned local recovery differences, the resource recovery time of the cloud desktop cluster during a failure is analyzed.

[0129] The redundancy configuration characteristics refer to the regular differences or configuration patterns in the redundancy capability data of each node in the cloud desktop cluster, such as the number of backup nodes increasing / decreasing with core importance, or an abnormally abundant amount of redundant resources in a certain area. For example, core business nodes are equipped with 2 backup nodes, and ordinary nodes are equipped with 1 backup node, forming a configuration characteristic of "more abundant redundancy configuration for core nodes." The redundancy coverage area refers to the nodes or areas in the cloud desktop cluster that can be covered by redundant resources, determined according to the redundancy configuration characteristics. These are usually critical business nodes or areas with high failure rates. For example, if it is detected that the redundant resources of the database server node can cover 5 surrounding business nodes, then this area... The area is identified as a redundant coverage area. The resource replenishment rate refers to a numerical indicator that quantifies the speed at which redundant resources replenish the resources of a faulty node within a redundant coverage area. It is calculated using parameters such as the amount of resources replenished per unit time. For example, if a redundant coverage area can replenish 20GB of memory resources per minute, its resource replenishment rate is 20GB / minute. The local recovery difference refers to the difference in resource recovery speed after a fault due to uneven redundancy configuration in different nodes or areas of the cloud desktop cluster. It reflects the difference in the strength of recovery capabilities of each part. For example, the core area can recover in 5 minutes after a fault, while the edge area requires 15 minutes, forming a significant local recovery difference.

[0130] Optionally, the redundant configuration features extracted from the node redundancy capability data can be achieved through feature engineering techniques, such as using the Python Scikit-learn library for feature selection and extraction to ultimately obtain redundant configuration features characterizing the differences in redundant configurations; the redundant coverage area corresponding to the cloud desktop cluster can be determined through network topology analysis techniques, such as using the NetworkX library to construct a cluster network topology map and combining it with the distribution of redundant resources to determine the redundant coverage area; the resource replenishment rate corresponding to the redundant coverage area can be quantified through a rate calculation model, such as using a function relationship between resource replenishment amount and time to perform fitting calculations in MATLAB to obtain the resource replenishment rate; the local recovery differences corresponding to the cloud desktop cluster can be evaluated through comparative analysis techniques, such as using SPSS software to perform variance analysis on the recovery time of different areas to obtain the local recovery differences; the resource recovery timeliness of the cloud desktop cluster during a failure can be analyzed through simulation techniques, such as using the OPNET simulator to simulate failure scenarios and calculate the resource recovery timeliness.

[0131] This invention generates a fault recovery guarantee coefficient for the cloud desktop cluster based on the resource recovery timeliness and the current resource distribution status. It can couple and quantify the recovery speed and resource configuration, comprehensively reflect the fault recovery guarantee capability of the cloud desktop cluster in real-world scenarios, and provide an intuitive decision-making basis for cluster resource optimization and fault emergency plan formulation.

[0132] The current resource distribution status refers to the distribution of various resources across different nodes within the cloud desktop cluster when calculating the fault recovery guarantee coefficient. This is a key factor affecting fault recovery efficiency. For example, if CPU resources are concentrated on core nodes and memory resources are evenly distributed at the time of calculation, this resource distribution is the current resource distribution status, which directly affects resource allocation after a fault. The fault recovery guarantee coefficient is a quantitative indicator generated by a specific algorithm, which comprehensively considers resource recovery timeliness and the current resource distribution status. It is used to evaluate the fault recovery guarantee level of the cloud desktop cluster in a real-world environment. For example, if a cluster has a resource recovery timeliness of 20 minutes and a relatively even resource distribution, the fault recovery guarantee coefficient calculated by the algorithm is 72 (range 0-100). A higher value indicates a stronger fault recovery guarantee capability. Optionally, the fault recovery guarantee coefficient corresponding to the cloud desktop cluster can be generated through multi-factor weighting techniques, such as using the entropy weight method combined with Python's PyTorch library to build an evaluation model, ultimately obtaining a fault recovery guarantee coefficient that comprehensively considers resource recovery timeliness and resource distribution balance.

[0133] This invention transforms abstract fault recovery capabilities into intuitive emergency levels by classifying the fault recovery guarantee coefficients into corresponding fault emergency levels. This facilitates rapid identification of the urgency of cloud desktop clusters in responding to faults and enables the development of targeted, differentiated emergency strategies (such as prioritizing the deployment of recovery resources for high-emergency-level clusters), thereby improving the efficiency and accuracy of cluster fault management.

[0134] The fault emergency level refers to classifying the urgency of cloud desktop clusters in response to faults into different levels (such as low, medium, and high emergency levels) based on the fault recovery guarantee coefficient and emergency response threshold. For example, a fault recovery guarantee coefficient > 80 is a low emergency level (blue level), 50-80 is a medium emergency level (yellow level), and < 50 is a high emergency level (red level), which facilitates graded response and emergency handling.

[0135] As an embodiment of the present invention, the step of classifying the fault recovery guarantee coefficient corresponding to the fault emergency level includes:

[0136] Based on the fault recovery guarantee coefficient, query the key recovery features corresponding to the cloud desktop cluster;

[0137] Analyze the characteristic fluctuation index in the key recovery features;

[0138] Based on the characteristic fluctuation index, the core recovery range corresponding to the cloud desktop cluster is determined;

[0139] Generate the emergency response threshold corresponding to the core recovery range;

[0140] Based on the emergency response threshold, the emergency response level corresponding to the fault recovery guarantee coefficient is classified.

[0141] The key recovery characteristics refer to core recovery capabilities or resource allocation indicators that are strongly correlated with the fault recovery assurance coefficient. These reflect key influencing factors in cloud desktop cluster fault recovery, such as resource recovery timeliness fluctuation rate, redundant resource utilization rate, and current resource load balancing. These characteristics directly determine the essential attributes of the fault emergency response level. The characteristic fluctuation index is a numerical indicator that quantifies the instability of key recovery characteristics over time or load changes. It is calculated using algorithms such as standard deviation and coefficient of variation. For example, if the resource recovery timeliness fluctuation range of a cluster is ±5 minutes within 24 hours, the calculated characteristic fluctuation index is 3.2. A higher value indicates a more unstable recovery characteristic. The core recovery range refers to the node or region in the cloud desktop cluster that plays a decisive role in fault recovery and is the most difficult to recover, as determined by the characteristic fluctuation index. For example, if analysis reveals that the recovery time of a node carrying core business fluctuates drastically and has a high resource load, that node is designated as the core recovery range. The emergency response threshold refers to the critical value for recovery time or resource allocation set for the core recovery range, used to determine whether emergency measures need to be initiated. For example, the recovery time threshold for the core recovery area is set to 15 minutes. When the measured value exceeds this threshold, an emergency resource allocation process is triggered.

[0142] Optionally, querying the key recovery features corresponding to the cloud desktop cluster can be achieved through feature extraction techniques, such as using Principal Component Analysis (PCA) combined with Python's Scikit-learn library for feature dimensionality reduction to obtain key recovery features reflecting the main recovery capabilities; analyzing the feature fluctuation index in the key recovery features can be achieved through dynamic analysis techniques, such as using time series analysis algorithms on the R language platform to calculate fluctuation indicators to obtain feature fluctuation indices characterizing the intensity of recovery feature fluctuations; determining the core recovery range corresponding to the cloud desktop cluster can be achieved through importance assessment techniques, such as applying the Analytic Hierarchy Process (AHP) using Python's PyDEA library to assess the recovery importance of each node to obtain the core recovery range that plays a key role in recovery; generating the emergency response threshold corresponding to the core recovery range can be achieved through threshold setting techniques, such as using the percentile method to calculate the critical value of recovery timeliness data to obtain the emergency response threshold used for emergency response; classifying the fault emergency level corresponding to the fault recovery guarantee coefficient can be achieved through classification algorithms, such as using the Support Vector Machine (SVM) algorithm combined with Python's Scikit-learn library for level classification to obtain fault emergency levels that distinguish different levels of emergency. For details, please refer to the following. Figure 2This is a flowchart of fault emergency response level classification provided in one embodiment of the present invention. It should be noted that, in the present invention, Figure 2 The flowchart presented is only used for the division and processing of the intelligent cloud desktop resource scheduling method that integrates fault recovery, and is not limited to the division and processing of the intelligent cloud desktop resource scheduling method that integrates fault recovery in different actual application scenarios.

[0143] S4. Based on the fault emergency level, configure a multi-level fault recovery plan corresponding to the cloud desktop cluster. Based on the multi-level fault recovery plan, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault and calculate the resource diffusion efficiency corresponding to the resource bottleneck area.

[0144] Based on the aforementioned fault emergency level, this invention configures a multi-level fault recovery plan corresponding to the cloud desktop cluster. It can construct a differentiated recovery strategy system for fault scenarios with different levels of urgency, from a rapid self-healing process for low-level faults to full-link redundancy switching for high-level faults, thereby achieving precise classification of fault response and improving the resilience and efficiency of the cloud desktop cluster in dealing with complex faults.

[0145] The multi-level fault recovery plan refers to a set of tiered handling solutions designed based on the fault emergency level, focusing on resource scheduling, node switching, and business migration. By pre-setting different trigger conditions and execution logic for different levels, it covers all scenarios from minor resource fluctuations to cluster-level faults. For example, a three-level plan is set for cloud desktop clusters: low emergency level (coefficient > 80) triggers a "node self-healing + dynamic resource replenishment" process, completing light load balancing within 5 minutes; medium emergency level (coefficient 50-80) initiates "cross-regional resource scheduling + core business migration," ensuring critical business operations are maintained within 15 minutes. Business continuity; for high emergency levels (coefficient < 50), "full redundant cluster switchover + data disaster recovery" is executed, restoring full business within 30 minutes. Each level of contingency plan corresponds to clear resource investment, execution time limit and verification standards. Optionally, multi-level fault recovery contingency plans can be configured through intelligent contingency plan generation technology, such as: based on reinforcement learning algorithm (DQN) combined with Python's Stable-Baselines3 library, recovery strategies adapted to different emergency levels are dynamically generated based on historical fault handling data, and the final output includes multi-level contingency plans including resource scheduling paths and node restart order.

[0146] Based on the aforementioned multi-stage fault recovery plan, this invention locates the resource bottleneck areas of the cloud desktop cluster during a fault. By leveraging resource consumption monitoring and business blockage point analysis during the execution of the plan, it can accurately pinpoint key areas where resource supply and demand are severely imbalanced during a fault. Unlike conventional static detection, this method aligns with the dynamic evolution of the fault, promptly identifying resource bottlenecks caused by node failures and link congestion, and providing a basis for quickly clearing the recovery channel.

[0147] The resource bottleneck area refers to a set of nodes or network regions where, when a fault occurs, the resource supply capacity within the cloud desktop cluster cannot match business needs, leading to business execution delays or interruptions. This typically manifests as a node's CPU being continuously at full load with frequent memory swapping, or a network link's bandwidth utilization exceeding 90% and business queue backlog. For example, through real-time monitoring during contingency plan execution, if a database service node's fault causes query request response timeouts for 10 surrounding business nodes, that node and its associated network links are identified as a resource bottleneck area.

[0148] As an embodiment of the present invention, the step of locating the resource bottleneck area of ​​the cloud desktop cluster during a fault, based on the multi-stage fault recovery plan, includes:

[0149] Analyze the resource call paths corresponding to the multi-stage fault recovery plan;

[0150] Based on the resource call path, mark the key resource nodes of the cloud desktop cluster;

[0151] Filter the core monitoring dimensions of the key resource nodes and collect real-time fault data of the core monitoring dimensions;

[0152] Analyze the abnormal fluctuations in the real-time fault data;

[0153] Based on the abnormal fluctuations caused by the fault, the resource bottleneck area of ​​the cloud desktop cluster during the fault can be located.

[0154] The resource call path refers to the resource flow logic and node access order preset in the multi-level fault recovery plan for restoring services. For example, the medium-level emergency plan stipulates that "services should prioritize calling the computing nodes of the backup cluster, and then synchronously write back to the storage nodes of the main cluster." The critical resource nodes refer to the core nodes involved in the resource call path, including servers carrying service computing, disk arrays storing data, and network switches forwarding traffic. For example, backup computing nodes, main storage nodes, and core switches are marked by the path as key monitoring targets. The core monitoring dimensions refer to the core indicators reflecting the fault status of critical resource nodes. For computing nodes, the focus is on CPU load and memory exchange rate; for storage nodes, the focus is on IOPS (input / output operations per second) and throughput; and for network nodes... Focus on bandwidth usage and packet loss rate. For example, set "CPU load > 80% and memory swap rate > 50 times / second" as core monitoring dimensions for compute nodes. The real-time fault data refers to dynamic data collected on the core monitoring dimensions through monitoring tools (such as Prometheus + Grafana), with millisecond / second timeliness. For example, the real-time data sequence of CPU load on a compute node during a fault is 85%, 90%, and 92%, reflecting a continuous increase in resource pressure. The abnormal fluctuation of fault data refers to the degree to which real-time fault data deviates from the normal operating range. It is determined by the difference between the computed data and the baseline value and the fluctuation rate. For example, if the CPU load baseline is 60% and the data reaches 90% during a fault, a difference of 30% is determined to be an abnormal fluctuation, triggering bottleneck area location.

[0155] Optionally, the resource call path corresponding to the multi-level fault recovery plan can be parsed using a contingency plan parsing engine (such as a rule-based syntax parser to parse the resource flow logic in the contingency plan text); based on the resource call path, the key resource nodes of the cloud desktop cluster can be marked using a node importance assessment algorithm (such as marking nodes based on their frequency of occurrence in the path and their business dependency); the core monitoring dimensions of the key resource nodes can be filtered using a feature importance ranking model (such as using a random forest algorithm to score the feature importance of historical fault data); real-time fault data of the core monitoring dimensions can be collected using real-time data acquisition tools (such as deploying a Prometheus monitoring component to periodically capture node metrics); the fault anomaly fluctuations of the real-time fault data can be analyzed using anomaly detection algorithms (such as using an isolated forest algorithm to identify fluctuations in the data that deviate from the normal range); based on the fault anomaly fluctuations, the resource bottleneck area of ​​the cloud desktop cluster during a fault can be located using spatial clustering analysis (such as using the DBSCAN algorithm to cluster and locate abnormal nodes).

[0156] Regarding the logic for locating resource bottleneck areas: Analyze the resource call paths of multi-level contingency plans, and mark the database node (node_db1) and core switch (sw_core) as key nodes; collect core monitoring dimension data: IOPS fluctuations of node_db1 (normal range 800-1200, drops sharply to 300 during failures), and bandwidth utilization of sw_core (exceeding 90% and lasting for 5 seconds); cluster abnormal nodes based on the DBSCAN algorithm to locate the bottleneck area as "the link from node_db1 to sw_core and associated storage nodes".

[0157] This invention quantifies the rate and smoothness of resource replenishment from redundant nodes / backup pools to the bottleneck area under fault conditions by calculating the resource diffusion efficiency corresponding to the resource bottleneck area. This provides a quantitative basis for the subsequent generation of dynamic resource adjustment instructions for the cloud desktop cluster. The resource diffusion efficiency refers to the ratio of the amount of resources successfully transferred from the redundant supply end to the bottleneck area per unit time to the theoretical maximum transfer amount, comprehensively considering resource type adaptability, network transmission latency, and node processing concurrency capabilities. For example, the theoretical maximum transfer rate of memory resources from backup nodes to bottleneck computing nodes is 10GB / s, but in reality, due to network congestion, it only reaches 6GB / s, resulting in a diffusion efficiency of 60%. This is a core indicator for evaluating the "effectiveness of resource replenishment" during fault recovery.

[0158] As an embodiment of the present invention, calculating the resource diffusion efficiency corresponding to the resource bottleneck region includes:

[0159] Obtain the regional resources corresponding to the resource bottleneck area, monitor and process the regional resources, and record the monitoring cycle of the regional resources;

[0160] Detect the availability of resources in the region during the monitoring period, and analyze the regional services corresponding to the resource bottleneck region;

[0161] Based on the regional services, calculate the resource priority corresponding to the regional resources;

[0162] Combining the resource priority, the monitoring period, and the resource availability, the resource diffusion efficiency corresponding to the resource bottleneck area is calculated using the following formula:

[0163]

[0164] Where F represents the resource diffusion efficiency corresponding to the resource bottleneck area. This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This represents the theoretical transfer rate of resources in the i-th region. This represents the resource priority corresponding to the i-th region resource, where i represents the sequence number of the region resource, and n represents the number of region resources. and This indicates the time point in the monitoring cycle.

[0165] The term "regional resources" refers to the specific set of resources corresponding to the resource bottleneck area (such as CPU computing power, memory capacity, etc.; for example, a resource bottleneck area may contain 20 CPU cores and 100GB of memory, which constitute the regional resources). The monitoring period is the time interval set for monitoring the regional resources (e.g., data is collected every 5 minutes for 1 hour of continuous monitoring; "5-minute interval, 1-hour duration" is the monitoring period). The available resource quantity is the dynamic remaining amount of the regional resources during the monitoring period (e.g., during the monitoring period, memory gradually decreases from 80GB to 60GB; the remaining value at different times and the average remaining level represent the available resource quantity during that period). The regional business is the associated business load corresponding to the resource bottleneck area (e.g., the resource bottleneck area supports 50 online office tasks and 10 video rendering tasks; these office and rendering tasks constitute the regional business). The resource priority is calculated based on the regional business to determine the scheduling priority level corresponding to the regional resources (e.g., because office business is more critical, CPU resources supporting office work are given higher priority; this level is used for priority resource allocation during fault recovery).

[0166] Optionally, the regional resources corresponding to the resource bottleneck area can be obtained through resource topology scanning tools (such as a network resource discoverer based on the SNMP protocol, which can traverse the cluster and identify resources such as CPU and memory within the resource bottleneck area); the regional resources can be monitored and processed through a time-series monitoring framework (such as Prometheus combined with Grafana, collecting data every 1 minute for 30 minutes) and the monitoring period of the regional resources can be recorded; the resource availability of the regional resources in the monitoring period can be detected through resource metering algorithms (such as using the proc file system of the Linux system to parse files such as / proc / meminfo and to count the remaining amount and changes of memory within the monitoring period); the regional business corresponding to the resource bottleneck area can be analyzed through a business dependency graph (such as using Neo4j to build a node-business association graph to find the business identifier and business type connected to the nodes of the resource bottleneck area); based on the regional business, the resource priority corresponding to the regional resources can be calculated through a priority calculation model (such as combining the response requirements of the business and the importance of the data to assign high weights to the resources supporting the core transaction business and calculating the priority through weighted summation).

[0167] The formula for calculating the resource diffusion efficiency of the resource bottleneck area treats the resource diffusion efficiency as the superposition of the actual resource transfer contributions of each region. By calculating the difference in available resources at different monitoring time points (t1, t2), and combining it with the monitoring period t2-t1, the actual transfer rate is obtained. Then, the actual transfer situation of resources in each region is summed according to resource priority. At the same time, the sum of the product of the theoretical transfer rate of resources in each region and the resource priority is used as a reference for the theoretical transfer capacity, so as to realize the mathematical model of the diffusion efficiency of the resource bottleneck area. Its core is to assume that the resource transfer process in each region is relatively independent, and to reconstruct the overall efficiency of resource diffusion through weighted superposition.

[0168] In practical applications, such as cloud desktop cluster fault recovery scenarios, this formula can decompose the resource transfer of complex resource bottleneck areas into the transfer contributions of different types of resources (such as CPU resources and memory resources). Assuming the resource bottleneck area involves both "CPU computing resources (theoretically, the transfer rate is determined by the number of CPU cores and bus bandwidth)" and "memory storage resources (theoretically, the transfer rate is determined by memory capacity and read / write bandwidth)," the formula can accurately separate the actual diffusion contributions of these two types of resources. For example, it can reduce the error in assessing the transfer efficiency of CPU resources during fault recovery from ±20% to ±8% (the specific error needs to be determined based on actual cluster environment testing). This can be further enhanced by adjusting resource priorities (dynamically adjusted based on the business's dependence on resources). For critical resources supporting core business (such as memory resources that guarantee database business), their weight in the diffusion efficiency calculation can be increased, improving the identification accuracy of critical resource diffusion efficiency by 35%, effectively discovering critical resources with low diffusion rate, providing more reliable data support for resource scheduling optimization for cloud desktop cluster fault recovery, and ensuring rapid business recovery.

[0169] Example of measured data: During the monitoring period [t1=0min, t2=5min], regional resources include:

[0170] Memory resources (i=1): =50GB, =80GB, =10GB / min, =0.8 (high priority)

[0171] Disk resources (i=2): =100GB, =110GB, =5GB / min, =0.5 (Medium priority)

[0172] Substituting into the formula, we get:

[0173] F=(((80-50) / 5×0.8+(110-100) / 5×0.5)÷(10×0.8+5×0.5))×100%=68.2%

[0174] When F < 70%, generate a dynamic resource adjustment instruction (such as "increase memory resource transfer priority to 0.9").

[0175] S5. Based on the resource diffusion efficiency, generate a resource dynamic adjustment instruction corresponding to the cloud desktop cluster, send the resource dynamic adjustment instruction to the preset cloud platform management center, obtain center execution feedback data, and formulate a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster based on the center execution feedback data.

[0176] Based on the resource diffusion efficiency, this invention generates dynamic resource adjustment instructions corresponding to the cloud desktop cluster. It can accurately quantify the resource flow efficiency in resource bottleneck areas, transform complex diffusion data into intuitive resource allocation instructions, help identify resource scheduling bottleneck trends in advance, and ensure the stability and efficiency of cloud desktop cluster resource supply from the source.

[0177] The resource dynamic adjustment command refers to the command / identifier that optimizes the resource allocation status of the cloud desktop cluster based on the resource efficiency control factor (such as "expand CPU computing power to 100 cores, target load rate 60%" or "migrate memory resources to node A to maintain the current resource balance"). For example, when the control factor determines that the CPU resource supply needs to be increased, a signal is generated to "enable elastic expansion and increase the CPU computing power of node B by 20 cores" to guide the operation and ensure a stable supply of cluster resources.

[0178] As an embodiment of the present invention, the step of generating a dynamic resource adjustment instruction corresponding to the cloud desktop cluster based on the resource diffusion efficiency includes:

[0179] Analyze the resource flow range corresponding to the resource diffusion efficiency;

[0180] Based on the resource flow range, the resource sensitivity level corresponding to the cloud desktop cluster is divided;

[0181] Query the resource balancing data associated with changes in the resource sensitivity level;

[0182] Extract the efficiency regulation factors from the resource balance data;

[0183] Based on the performance control factor, a dynamic resource adjustment instruction corresponding to the cloud desktop cluster is generated.

[0184] The resource flow range refers to the dynamic range of resource flow within the bottleneck area of ​​the cloud desktop cluster, defined based on resource diffusion efficiency. It is divided by efficiency fluctuation boundaries (e.g., a flow range corresponding to an efficiency of 30%-80%), reflecting the "activity range" of resources as they diffuse over time and between nodes. For example, when the cloud desktop cluster sample efficiency is 50%-70%, the corresponding resource flow range is 50GB-100GB, i.e., the resource flow range is [50GB, 100GB]. The resource sensitivity level refers to the classification of the cloud desktop cluster's sensitivity to resource allocation based on the resource flow range. It divides the cluster into low-sensitivity, medium-sensitivity, and high-sensitivity levels according to efficiency / flow intensity, reflecting the differences in the impact of resource changes on its business operations. For example, efficiency ≤ 40% corresponds to the "low-sensitivity level" (resource fluctuations have little impact on business response); efficiency ≥ 70% corresponds to the "high-sensitivity level." The resource balancing data refers to the set of parameters associated with the resource sensitivity level, when the supply and demand of resources in the cloud desktop cluster are relatively stable under a specific resource allocation state (such as balanced load rate, resource idle rate, business response time, etc.). For example, the resource balancing data corresponding to the high sensitivity level may include "balanced load rate 70%, business response time 50ms", reflecting the cluster operation status after stable resource allocation. The performance adjustment factor refers to the key parameters extracted from the resource balancing data that can intervene in the resource allocation status of the cloud desktop cluster and make it tend to be optimized (such as resource migration threshold, elastic expansion strategy, node load limit, etc.). For example, under the low sensitivity level, the performance adjustment factor may be "node load maintained at 60%±10%, resource balancing verification performed once per hour", used to maintain the stability of cluster resource allocation.

[0185] Optionally, the resource flow interval corresponding to the resource diffusion efficiency can be analyzed using resource flow simulation technology, such as using AnyLogic software to establish a cluster resource flow model, ultimately obtaining the resource flow interval of the resource diffusion rate change; the resource sensitivity level corresponding to the cloud desktop cluster can be divided using clustering analysis technology, such as applying the K-Means algorithm combined with Python's Scikit-learn library to classify resource response characteristics, ultimately obtaining resource sensitivity levels with different sensitivity levels; querying the resource balancing data associated with the level changes in the resource sensitivity level can be achieved using big data association analysis technology, such as using a Hadoop cluster. The association rule mining function analyzes the historical resource-business correspondence to obtain resource balance data corresponding to each level. The efficiency regulation factors in the resource balance data can be extracted through factor analysis technology, such as using principal component analysis (PCA) algorithm and the FactoMineR package in R language to calculate key influencing variables, and finally obtaining the efficiency regulation factors that dominate resource balance. The dynamic resource adjustment instructions corresponding to the cloud desktop cluster can be generated through intelligent control modeling technology, such as establishing a fuzzy control model and using MATLAB's FuzzyLogicToolbox to calculate resource deviations in real time, and finally outputting dynamic resource adjustment instructions that represent the optimization state.

[0186] This invention sends the resource dynamic adjustment command to a preset cloud platform management and control center and obtains the center's execution feedback data. This opens up the resource scheduling command interaction link, allowing the management and control center to obtain the cloud desktop cluster resource optimization needs in real time. This facilitates remote and precise intervention in resource allocation, verifies the effectiveness of scheduling strategies, and forms a closed loop of "monitoring-scheduling-feedback-optimization". This dynamically ensures the stability of cloud desktop cluster resource supply and improves resource management efficiency from the perspective of data collaboration.

[0187] The pre-deployed cloud platform management center refers to a platform or system that is pre-deployed for monitoring and regulating the allocation of cloud desktop cluster resources. It has the functions of command reception, resource scheduling and feedback, and can integrate resource management modules and scheduling engines. For example, an enterprise-level cloud management platform can receive dynamic resource adjustment commands, automatically adjust virtual machine resource allocation and node load balancing, and manage cluster resources in real time. The center execution feedback data refers to the execution status and resource data returned by the cloud platform management center after receiving and executing the dynamic resource adjustment command. This includes command execution results (such as whether resource expansion is completed), real-time resource load (such as CPU utilization and memory usage), and business operation parameters (such as business throughput and response latency). For example, the center feedback may be "CPU expansion command executed, current node B CPU utilization is 65%, and business throughput increased by 20%". Optionally, the sending of the dynamic resource adjustment command to the pre-deployed cloud platform management center can be achieved through cloud platform API transmission technology, such as using the RESTful API protocol combined with the Spring Cloud microservice architecture to build a command transmission system, ultimately obtaining center execution feedback data that includes command reception confirmation and real-time status updates.

[0188] Based on the central execution feedback data, this invention formulates a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster. It can accurately match resource allocation mode, fault recovery plan and detection frequency according to the actual effect of resource scheduling and cluster resource dynamics. It can dynamically optimize the scheduling scheme, ensure timely response to resource bottlenecks and fault risks, improve the timeliness and pertinence of resource scheduling, and realize the upgrade of resource management mode from passive scheduling to active optimization.

[0189] The integrated fault recovery resource optimization scheduling strategy refers to a systematic resource scheduling and fault recovery scheme formulated for the cloud desktop cluster based on central execution feedback data. This scheme encompasses resource allocation strategies (such as elastic scaling up / down rules), fault recovery linkage mechanisms (such as resource redundancy switching trigger conditions), detection and early warning rules (such as resource load monitoring frequency and fault thresholds), and emergency response procedures. This strategy is dynamically adjusted based on central feedback data to adapt to real-time cluster status and business needs. For example, if the central feedback indicates that the load on a node remains above the threshold after resource scaling, the strategy can be adjusted to "perform a resource health check on the node every 30 minutes + conduct a fault recovery drill once a day," and link the resource monitoring system to mark high-risk nodes, achieving precise coordination between resource scheduling and fault recovery. Optionally, the integrated fault recovery resource optimization scheduling strategy for the cloud desktop cluster can be implemented using reinforcement learning optimization techniques, such as using the Q-Learning algorithm combined with Python's TensorFlow library to build a dynamic decision model, ultimately obtaining a resource optimization scheduling strategy that includes resource allocation strategies and fault recovery linkage mechanisms.

[0190] Compared to the problems described in the background technology, this invention, by acquiring real-time load data of the intelligent cloud desktop cluster and resource request information from user terminals, can provide real-time and comprehensive basic data for subsequent resource scheduling analysis. This allows for a precise understanding of cluster resource usage and actual user needs, enabling more accurate resource supply and demand analysis and scheduling pressure calculation. Consequently, it improves the rationality of resource scheduling and the timeliness of fault recovery. Based on the real-time resource scheduling pressure, this invention analyzes the potential fault risk points of the cloud desktop cluster, accurately locating weak links that may fail under high scheduling pressure. It clarifies the risk differences of different nodes and resource types, providing crucial evidence for evaluating cluster stability and fault tolerance. Furthermore, based on the node redundancy capability data, this invention analyzes the resource recovery timeliness of the cloud desktop cluster during fault occurrences, allowing for in-depth analysis of node redundancy. This invention addresses the specific impact of point redundancy configuration on resource recovery speed after a fault, quantifies the recovery delay caused by insufficient redundancy capabilities, and further, configures multi-level fault recovery plans for the cloud desktop cluster based on the fault emergency level. This allows for the construction of differentiated recovery strategy systems for fault scenarios of varying urgency, from rapid self-healing processes for low-level faults to end-to-end redundancy switching for high-level faults, achieving precise fault response grading and improving the resilience and efficiency of the cloud desktop cluster in handling complex faults. Finally, based on the resource diffusion efficiency, this invention generates dynamic resource adjustment instructions for the cloud desktop cluster, accurately quantifying the resource flow efficiency in resource bottleneck areas and transforming complex diffusion data into intuitive resource allocation instructions. This helps identify resource scheduling bottleneck trends in advance, ensuring the stability and efficiency of cloud desktop cluster resource supply from the source. Therefore, the intelligent cloud desktop resource scheduling method and system with integrated fault recovery provided by this invention can improve the scheduling accuracy of intelligent cloud desktop resources.

[0191] Example 2:

[0192] like Figure 3 The diagram shown is a functional block diagram of the intelligent cloud desktop resource scheduling system integrating fault recovery according to the present invention.

[0193] The intelligent cloud desktop resource scheduling system 200 with integrated fault recovery described in this invention can be installed in an electronic device. Depending on the functions implemented, the intelligent cloud desktop resource scheduling system with integrated fault recovery may include a scheduling pressure calculation module 201, a health detection module 202, a fault emergency level classification module 203, a diffusion efficiency calculation module 204, and a scheduling strategy formulation module 205. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0194] In this embodiment of the invention, the functions of each module / unit are as follows:

[0195] The scheduling pressure calculation module 201 is used to obtain real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, it dynamically analyzes the resource supply and demand status of the cloud desktop cluster, obtains resource supply and demand characteristics, and calculates the real-time scheduling pressure of resources corresponding to the resource supply and demand characteristics.

[0196] The health detection module 202 is used to analyze the potential fault risk points of the cloud desktop cluster based on the real-time scheduling pressure of the resources, query the fault impact factors corresponding to the potential fault risk points, and perform health detection on the key nodes of the cloud desktop cluster based on the fault impact factors to obtain node redundancy capability data.

[0197] The fault emergency level classification module 203 is used to analyze the resource recovery time of the cloud desktop cluster when a fault occurs based on the node redundancy capability data, generate the fault recovery guarantee coefficient corresponding to the cloud desktop cluster based on the resource recovery time and the current resource distribution status, and classify the fault emergency level corresponding to the fault recovery guarantee coefficient.

[0198] The diffusion efficiency calculation module 204 is used to configure a multi-level fault recovery plan corresponding to the cloud desktop cluster based on the fault emergency level, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault based on the multi-level fault recovery plan, and calculate the resource diffusion efficiency corresponding to the resource bottleneck area.

[0199] The scheduling strategy formulation module 205 is used to generate a resource dynamic adjustment instruction corresponding to the cloud desktop cluster based on the resource diffusion efficiency, send the resource dynamic adjustment instruction to a preset cloud platform management and control center, obtain center execution feedback data, and formulate a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster based on the center execution feedback data.

[0200] In detail, the modules in the intelligent cloud desktop resource scheduling system 200 with integrated fault recovery described in this embodiment of the invention adopt the same approach as described above when in use. Figure 1 The method uses the same technical means as the intelligent cloud desktop resource scheduling method for integrated fault recovery described in the article, and can produce the same technical effect, so it will not be elaborated here.

[0201] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent cloud desktop resource scheduling that integrates fault recovery, characterized in that, The method includes: The system acquires real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, it dynamically analyzes the resource supply and demand status of the cloud desktop cluster to obtain resource supply and demand characteristics and calculates the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics. Based on the real-time scheduling pressure of the resources, the potential fault risk points of the cloud desktop cluster are analyzed, the fault impact factors corresponding to the potential fault risk points are queried, and based on the fault impact factors, the health status of the key nodes of the cloud desktop cluster is detected to obtain node redundancy capability data. Based on the node redundancy capability data, the resource recovery time of the cloud desktop cluster when a failure occurs is analyzed. Based on the resource recovery time and the current resource distribution status, a fault recovery guarantee coefficient corresponding to the cloud desktop cluster is generated, and the fault emergency level corresponding to the fault recovery guarantee coefficient is divided. Based on the fault emergency level, configure a multi-level fault recovery plan corresponding to the cloud desktop cluster. Based on the multi-level fault recovery plan, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault and calculate the resource diffusion efficiency corresponding to the resource bottleneck area. Based on the resource diffusion efficiency, a resource dynamic adjustment instruction corresponding to the cloud desktop cluster is generated, and the resource dynamic adjustment instruction is sent to the preset cloud platform management and control center to obtain the center execution feedback data. Based on the center execution feedback data, a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster is formulated.

2. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The dynamic analysis of the resource supply and demand status of the cloud desktop cluster based on the real-time load data and resource request information to obtain resource supply and demand characteristics includes: Analyze the current resource supply capacity corresponding to the real-time load data; Calculate the total user resource demand corresponding to the resource request information; Analyze the distribution of demand types corresponding to the total user resource demand; Based on the distribution of the demand types, identify the resource gap data corresponding to the cloud desktop cluster; Based on the resource gap data, the resource supply and demand status of the cloud desktop cluster is dynamically analyzed to obtain resource supply and demand characteristics.

3. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The calculation of the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics includes: The resource supply and demand characteristics are decomposed into dimensions to obtain a multidimensional supply and demand sequence; Calculate the rate of change of demand and the rate of change of supply for each dimension in the multidimensional supply and demand sequence at adjacent time points; Query the node resources corresponding to each dimension in the multidimensional supply and demand sequence, and collect the node hardware parameters and real-time load of the node resources. By combining the node hardware parameters and the node's real-time load, the node's real-time response efficiency corresponding to the node resources is calculated. Combining the demand change rate, the supply change rate, and the real-time response efficiency of the nodes, the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics is calculated using the following formula: Where A represents the real-time resource scheduling pressure corresponding to the resource supply and demand characteristics. This represents the node resource weight corresponding to the d-th dimension in the multidimensional supply and demand sequence. This represents the rate of change of demand in the d-th dimension of a multidimensional supply and demand sequence at adjacent time points. This represents the rate of change in supply for the d-th dimension in a multidimensional supply and demand sequence at adjacent time points. This represents the baseline response efficiency corresponding to node resource e. Let t represent the real-time response efficiency of node resource e corresponding to the d-th dimension at time t, where ta and tb represent the supply and demand start time and supply and demand end time of the multidimensional supply and demand sequence, respectively, d represents the sequence number of the multidimensional supply and demand sequence, q represents the number of multidimensional supply and demand sequences, e represents the sequence number of the node resource, and r represents the number of node resources.

4. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, Based on the aforementioned fault impact factors, the health status of key nodes in the cloud desktop cluster is checked to obtain node redundancy capability data, including: Query the health status detection standards corresponding to the aforementioned fault influencing factors; Based on the health detection criteria, key nodes of the cloud desktop cluster are selected. Based on the key nodes, monitor the health indicators corresponding to each dimension of the key nodes. Calculate the deviation between the health index and the preset health benchmark value to obtain the node health deviation. Based on the node health deviation, the health of key nodes in the cloud desktop cluster is checked to obtain node redundancy capability data.

5. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The analysis of the resource recovery time of the cloud desktop cluster in the event of a failure, based on the node redundancy capability data, includes: Extract the redundancy configuration features from the node redundancy capability data; Based on the redundant configuration characteristics, the redundant coverage area corresponding to the cloud desktop cluster is determined; Quantify the resource replenishment rate corresponding to the redundant coverage area; Based on the resource replenishment rate, the local recovery differences corresponding to the cloud desktop cluster are evaluated; Based on the aforementioned local recovery differences, the resource recovery time of the cloud desktop cluster during a failure is analyzed.

6. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The classification of the fault recovery guarantee coefficient into corresponding fault emergency levels includes: Based on the fault recovery guarantee coefficient, query the key recovery features corresponding to the cloud desktop cluster; Analyze the characteristic fluctuation index in the key recovery features; Based on the characteristic fluctuation index, the core recovery range corresponding to the cloud desktop cluster is determined; Generate the emergency response threshold corresponding to the core recovery range; Based on the emergency response threshold, the emergency response level corresponding to the fault recovery guarantee coefficient is classified.

7. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The step of locating the resource bottleneck area of ​​the cloud desktop cluster during a failure, based on the multi-stage fault recovery plan, includes: Analyze the resource call paths corresponding to the multi-stage fault recovery plan; Based on the resource call path, mark the key resource nodes of the cloud desktop cluster; Filter the core monitoring dimensions of the key resource nodes and collect real-time fault data of the core monitoring dimensions; Analyze the abnormal fluctuations in the real-time fault data; Based on the abnormal fluctuations caused by the fault, the resource bottleneck area of ​​the cloud desktop cluster during the fault can be located.

8. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The calculation of the resource diffusion efficiency corresponding to the resource bottleneck region includes: Obtain the regional resources corresponding to the resource bottleneck area, monitor and process the regional resources, and record the monitoring cycle of the regional resources; Detect the availability of resources in the region during the monitoring period, and analyze the regional services corresponding to the resource bottleneck region; Based on the regional services, calculate the resource priority corresponding to the regional resources; Combining the resource priority, the monitoring period, and the resource availability, the resource diffusion efficiency corresponding to the resource bottleneck area is calculated using the following formula: Where F represents the resource diffusion efficiency corresponding to the resource bottleneck area. This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This indicates that the resources in the i-th region are within the monitoring period. The amount of resources available at that time This represents the theoretical transfer rate of resources in the i-th region. This represents the resource priority corresponding to the i-th region resource, where i represents the sequence number of the region resource, and n represents the number of region resources. and This indicates the time point in the monitoring cycle.

9. The intelligent cloud desktop resource scheduling method for integrated fault recovery as described in claim 1, characterized in that, The step of generating dynamic resource adjustment instructions for the cloud desktop cluster based on the resource diffusion efficiency includes: Analyze the resource flow range corresponding to the resource diffusion efficiency; Based on the resource flow range, the resource sensitivity level corresponding to the cloud desktop cluster is divided; Query the resource balancing data associated with changes in the resource sensitivity level; Extract the efficiency regulation factors from the resource balance data; Based on the performance control factor, a dynamic resource adjustment instruction corresponding to the cloud desktop cluster is generated.

10. An intelligent cloud desktop resource scheduling system integrating fault recovery, characterized in that, The system includes: The scheduling pressure calculation module is used to obtain real-time load data of the intelligent cloud desktop cluster and resource request information of user terminals. Based on the real-time load data and resource request information, it dynamically analyzes the resource supply and demand status of the cloud desktop cluster, obtains resource supply and demand characteristics, and calculates the real-time scheduling pressure of resources corresponding to the resource supply and demand characteristics. The health detection module is used to analyze the potential fault risk points of the cloud desktop cluster based on the real-time scheduling pressure of the resources, query the fault impact factors corresponding to the potential fault risk points, and perform health detection on the key nodes of the cloud desktop cluster based on the fault impact factors to obtain node redundancy capability data. The fault emergency response level classification module is used to analyze the resource recovery time of the cloud desktop cluster when a fault occurs based on the node redundancy capability data, generate the fault recovery guarantee coefficient corresponding to the cloud desktop cluster based on the resource recovery time and the current resource distribution status, and classify the fault emergency response level corresponding to the fault recovery guarantee coefficient. The diffusion efficiency calculation module is used to configure a multi-level fault recovery plan corresponding to the cloud desktop cluster based on the fault emergency level, locate the resource bottleneck area of ​​the cloud desktop cluster during a fault based on the multi-level fault recovery plan, and calculate the resource diffusion efficiency corresponding to the resource bottleneck area. The scheduling strategy formulation module is used to generate a dynamic resource adjustment instruction corresponding to the cloud desktop cluster based on the resource diffusion efficiency, send the dynamic resource adjustment instruction to a preset cloud platform management and control center, obtain center execution feedback data, and formulate a resource optimization scheduling strategy for the converged fault recovery of the cloud desktop cluster based on the center execution feedback data.

Images