Dynamic allocation management system for monitoring computing power resources in real time

Abstract

The present invention relates to the technical field of computing power resource management. Disclosed is a dynamic allocation management system for monitoring computing power resources in real time, which system is used for solving the problem of it not being possible for existing dynamic allocation management systems for computing power resources to perform efficient matching, optimal utilization, accurate evaluation and dynamic allocation on heterogeneous computing power resources. The system comprises a central cloud and several edge nodes in communication connection with the central cloud, wherein the central cloud is internally provided with a real-time monitoring and prediction unit, a dynamic allocation management unit and a fault detection and processing unit. In the present invention, by means of analyzing the utilization rate and latency information of heterogeneous computing power resources in detail, resources most suitable for a target processing task can be efficiently matched; by means of eliminating computing power resources with lower latency scores, the overall resource utilization efficiency is improved, and resource waste is avoided; and resource demand prediction is performed in combination with a plurality of parameters, such that the performance and value of computing power resources to be allocated can be evaluated more accurately. Therefore, the effective management and optimal utilization of heterogeneous computing power resources are realized.

Description

A dynamic allocation and management system for real-time monitoring of computing resources Technical Field

[0001] This invention relates to the field of computing resource management technology, specifically to a dynamic allocation management system for real-time monitoring of computing resources. Background Technology

[0002] With the rapid development of cloud computing, big data, and other industries in China, the demand for computing resources is also increasing. To meet this demand, some large internet companies, research institutions, and cloud computing service providers have begun to invest resources in system development. Currently, some mature systems have been applied in real-world scenarios in China. These systems can monitor the real-time status of computing resources, such as the utilization rate and load of heterogeneous computing resources like CPUs, GPUs, FPGAs, and NPUs, and dynamically adjust resource allocation based on this information.

[0003] However, existing dynamic allocation management systems for computing resources are unable to efficiently match, optimize, accurately evaluate, and dynamically allocate heterogeneous computing resources, resulting in the inability to make better use of computing resources. Summary of the Invention

[0004] The purpose of this invention is to solve the problem that existing dynamic allocation management systems for computing resources cannot efficiently match, optimize, accurately evaluate, and dynamically allocate heterogeneous computing resources, and to propose a dynamic allocation management system for real-time monitoring of computing resources.

[0005] The objective of this invention can be achieved through the following technical solution: It includes a central cloud and several edge nodes connected to the central cloud. The central cloud is equipped with a real-time monitoring and prediction unit and a dynamic allocation management unit. The real-time monitoring and prediction unit is used to collect and analyze the status information of the heterogeneous computing resources corresponding to the edge nodes in real time using monitoring tools and algorithms. The specific analysis process is as follows:

[0006] The system acquires the target processing task, matches the required heterogeneous computing resources for that task, and determines the utilization rate of these resources. If the utilization rate is less than a set threshold, the resource is marked as a pending allocation resource and assigned the number i. The system then acquires the status information of the pending allocation resources, analyzes all latency durations within a preset time period to obtain the latency score, and acquires the corresponding latency threshold. The latency score is compared with the threshold; if the latency score is less than or equal to the threshold, the resource is removed; otherwise, the resource is reclassified. The system matches the computing power resource to be allocated with a set of preset models, each model having a corresponding model value. If the model of the computing power resource to be allocated matches a preset model, the model value of the matched preset model is marked as the model value XHi of the computing power resource to be allocated. The energy efficiency ratio NXI is obtained by dividing the number of floating-point operations of the computing power resource to be allocated by the power consumption. The utilization rate, number of load tasks, and energy consumption ratio of the computing power resource to be allocated are marked as LYi, FZi, and NXI, respectively. The system then inputs these values ​​into the resource demand prediction model to output the resource allocation prediction value of the computing power resource to be allocated. Finally, the resource allocation prediction value of the computing power resource to be allocated is sent to the dynamic allocation management unit.

[0007] The dynamic allocation management unit is used to dynamically allocate and manage heterogeneous computing resources.

[0008] In a preferred embodiment of the present invention, heterogeneous computing resources include CPU, GPU, FPGA, NPU and ASIC; status information includes performance data, utilization rate, number of load tasks and energy consumption data; performance data includes floating-point operation count, latency, model and temperature; energy consumption data includes power and energy efficiency ratio.

[0009] As a preferred embodiment of the present invention, the specific process for obtaining the extension score of the computing power resources to be allocated is as follows:

[0010] The delay durations are sorted according to the order of data collection. Adjacent delay durations are compared; if the delay duration collected earlier is greater than the delay duration collected later, the difference between the two is calculated to obtain the delay drop value. All delay drop values ​​are summed to obtain the total delay value. The average delay value is calculated by taking the average of all delay duration values. The maximum and minimum delay values ​​are then identified and labeled as the maximum and minimum delay durations, respectively. The total delay value, average delay value, maximum delay value, and minimum delay value are normalized and labeled as TZ, TP, TMAX, and TMIN, respectively. These values ​​are then substituted into the preset delay analysis model.

[0011] Output the delay score TYi of the computing resources to be allocated, where f1, f2, f3 and f4 are the weighting factors corresponding to the total delay, average delay, maximum delay and minimum delay, respectively.

[0012] In a preferred embodiment of the present invention, the resource demand forecasting model is specifically as follows: Output the pre-allocated computing power resources ZFi; where hy1, hy2, and hy3 are model thresholds; For extended weighting factors, Energy efficiency ratio weighting factor; Utilization rate weighting factor; Energy consumption ratio weighting factor; The weighting factor for the number of load tasks is i = 1, 2, ..., n, where n is the maximum value of the number.

[0013] As a preferred embodiment of the present invention, the specific process of dynamic allocation management is as follows:

[0014] Heterogeneous computing resources are sorted from largest to smallest according to their pre-defined resource allocation values. If the number of target processing tasks is less than or equal to the number of heterogeneous computing resources, a preset number of heterogeneous computing resources are selected sequentially from the top of the sort. The target processing tasks are then equally distributed and sent to the edge nodes corresponding to the heterogeneous computing resources, where they are processed by the corresponding heterogeneous computing resources within the edge nodes. If the number of target processing tasks is greater than the number of heterogeneous computing resources, the target processing tasks are sent to the edge node corresponding to the first-ranked heterogeneous computing resource until its utilization rate equals the maximum utilization rate threshold, at which point the sending stops. Then, the tasks are sent to the edge node corresponding to the second-ranked heterogeneous computing resource, and so on, until all target processing tasks have been distributed.

[0015] In a preferred embodiment of the present invention, a fault detection and processing unit is further provided within the central cloud; the fault detection and processing unit is used to perform fault detection and analysis on heterogeneous computing resources, and the specific process is as follows:

[0016] Acquire all temperatures corresponding to heterogeneous computing resources, compare each temperature with its corresponding preset maximum temperature, and count the number of temperatures equal to or higher than the maximum temperature to obtain the heterogeneous temperature count. Calculate the average temperature by averaging all temperatures. When the average temperature exceeds a set average threshold, acquire the voltage Uj and current Ij of the heterogeneous computing resources, j = 1, 2, ..., m, where m represents the acquisition time number and m represents the maximum value of that number. Substitute these values ​​into the formula... Output voltage fluctuation value SU and current fluctuation value SI; normalize the number of temperature variations, voltage fluctuation value, average temperature value, and current fluctuation value, and take the values ​​of the four, then substitute them into the formula. Obtain the fault detection value GYi of heterogeneous computing power resources; where β1, β2, β3 and β4 are the weighting factors corresponding to the number of abnormal temperatures YW, the average temperature value PJ, the voltage fluctuation value SU and the current fluctuation value SI, respectively.

[0017] Define two intervals: Interval 1 and Interval 2. When the fault detection value falls within Interval 1, shut down the edge node corresponding to the heterogeneous computing resource. Then, convert the fault detection value into a shutdown duration according to a certain ratio, execute the shutdown duration, and then restart. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. When the fault detection value falls within Interval 2, execute a self-check and repair program on the edge node corresponding to the heterogeneous computing resource. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. After the self-check and repair program is completed, recalculate the fault detection value of the heterogeneous computing resource. If the fault detection value still falls within Interval 2, mark the edge node corresponding to the heterogeneous computing resource as a fault node. Send the coordinates of the fault node, the name of the heterogeneous computing resource, and the fault detection value to the smart terminal of the corresponding maintenance personnel. The maintenance personnel will then go to the location of the fault node to repair or replace the heterogeneous computing resource.

[0018] Compared with the prior art, the beneficial effects of the present invention are:

[0019] 1. This invention, through detailed analysis of the utilization rate and latency information of heterogeneous computing resources, can efficiently match the resources most suitable for the target processing task; by eliminating computing resources with low latency scores, it improves the overall resource utilization efficiency and avoids resource waste; by combining multiple parameters (such as model, energy efficiency ratio, utilization rate, number of load tasks, and energy consumption ratio) for resource demand prediction, it can more accurately assess the performance and value of computing resources to be allocated. This achieves effective management and optimized utilization of heterogeneous computing resources.

[0020] 2. By monitoring key parameters such as temperature, voltage, and current of heterogeneous computing resources in real time, potential faults can be detected and addressed promptly, improving system stability and reliability. Different handling measures are automatically implemented based on different fault detection ranges, achieving intelligent and automated fault handling. This reduces service interruptions and data loss caused by faults, improving system availability. Attached Figure Description

[0021] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0022] Figure 1 is an overall principle block diagram of the present invention;

[0023] Figure 2 is a schematic diagram of the internal principle of the central cloud of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1

[0026] Please refer to Figure 1. A dynamic allocation and management system for real-time monitoring computing resources includes a central cloud and several edge nodes (edge ​​nodes are road inspection vehicles or other intelligent terminals with edge computing hosts) that are connected to the central cloud. The central cloud is equipped with a real-time monitoring and prediction unit, a dynamic allocation and management unit, and a fault detection and processing unit.

[0027] The real-time monitoring and prediction unit utilizes monitoring tools and algorithms to collect and analyze the status information of heterogeneous computing resources corresponding to edge nodes in real time. These heterogeneous computing resources include CPUs, GPUs, FPGAs, NPUs, and ASICs. A GPU (Graphics Processing Unit) is a microprocessor specifically designed for processing images and graphics. It is similar to a computer's central processing unit (CPU) but is specifically designed for processing image and graphics data. GPUs typically contain thousands of streaming processors, which can process large amounts of data in parallel, thus achieving efficient graphics rendering and computation. NVIDIA is a representative GPU manufacturer. An NPU (Neural Processing Unit) is a chip specifically designed for deep learning computation. It is one of the hottest technologies in the field of artificial intelligence in recent years and is widely used in various AI applications, such as autonomous driving, facial recognition, intelligent voice, and video surveillance. Through massively parallel processing units and efficient interconnect structures, NPUs accelerate complex calculations in deep neural networks, offering advantages such as high efficiency, low power consumption, low latency, stability, and programmability. FPGA: A flexible and programmable hardware platform with high computing performance and customizability, capable of accelerating and optimizing AI algorithms. In AI applications, it can be used to implement neural network accelerators, high-performance computing units, etc., providing high-performance and low-latency computing capabilities for computationally intensive AI tasks. The Intel Stratix 10NX FPGA is specifically designed for AI, featuring AI tensor blocks containing dense arrays of low-precision multipliers, tuned for matrix and vector multiplication, and capable of performing INT4, INT8, Block FP12, or Block FP16 operations. Furthermore, these tensor blocks can be cascaded to support large matrices. DPU: A Data Processing Unit (DPU) is a dedicated processor built around data. Due to the slowdown in CPU performance growth, DPUs emerged to seek more efficient computing chips. It employs specific software-defined technologies to support the virtualization of infrastructure-level resources, supporting functions such as storage, security, and quality of service management. The emergence of DPUs aims to efficiently handle large-scale data workloads within data centers, including tasks such as data transmission, reduction, protection, compression, analysis, and encryption.

[0028] Status information includes performance data, utilization rate, number of load tasks, energy consumption data, etc.; among which, performance data includes floating-point operation count, latency, model and temperature; energy consumption data includes power and energy efficiency ratio.

[0029] The specific analysis process of the real-time monitoring and prediction unit is as follows:

[0030] The system acquires the target processing task, matches the required heterogeneous computing resources for that task, and determines the utilization rate of these resources. If the utilization rate is less than a set threshold, the resource is marked as a pending allocation resource and assigned a number i, where i = 1, 2, ..., n, and n is the maximum number. The system then acquires the status information of the pending allocation resources, analyzes all latency durations within a preset time period, sorts the latency durations according to the order of acquisition, and compares adjacent latency durations. If the latency duration ranked earlier is greater than the latency duration ranked later... For delay duration, subtract the delay duration from the delay duration in the order of priority to obtain the delay difference value. Sum all delay difference values ​​to obtain the total delay value. Calculate the average delay value from all delay duration values. Obtain the maximum and minimum delay values ​​and label them as the maximum delay duration and minimum delay duration, respectively. Normalize the total delay value, average delay value, maximum delay duration, and minimum delay duration, and label the values ​​of these four values ​​as TZ, TP, TMAX, and TMIN, respectively. Substitute them into the preset delay analysis model:

[0031] Output the delay score TYi of the computing power resource to be allocated, where f1, f2, f3, and f4 are the weight factors corresponding to the total time drop, average delay, maximum delay, and minimum delay, respectively, and their values ​​are customizable, such as 1.31, 1.51, 1.67, and 1.75. Obtain the delay score threshold corresponding to the computing power resource to be allocated, and compare the delay score with the corresponding delay score threshold. If the delay score is less than or equal to the delay score threshold, the computing power resource to be allocated is removed; if the delay score is greater than the delay score threshold, the computing power resource to be allocated is added. The model is matched with several preset models, each with a corresponding model value. If the model of the computing resource to be allocated matches a preset model, the model value of the matched preset model is marked as the model value XHi of the computing resource to be allocated. The energy efficiency ratio NXI is obtained by dividing the number of floating-point operations of the computing resource to be allocated by the power value. The utilization rate, number of load tasks, and energy consumption ratio of the computing resource to be allocated are marked as LYi, FZi, and NXI, respectively. These values ​​are then substituted into the resource demand prediction model. Output the pre-allocated computing power resources ZFi; where hy1, hy2, and hy3 are model thresholds, and hy1 <hy2<hy3; For extended weighting factors, and As the energy efficiency ratio weighting factor, and As a utilization rate weighting factor, and As the energy consumption ratio weighting factor, and This is a weighting factor for the number of workload tasks; the pre-defined value of the computing resources to be allocated is sent to the dynamic allocation management unit.

[0032] Based on task requirements, suitable heterogeneous computing resources are selected, and their utilization rates are checked. Resources below a set threshold are marked as pending allocation resources and assigned numbers. The latency of these pending resources is collected, and the total latency, average latency, maximum latency, and minimum latency are calculated and normalized. The normalized values ​​are then substituted into a preset latency analysis model to calculate the latency score for each pending resource. The latency score is compared with the latency threshold, and resources that do not meet the requirements are eliminated. Preset models are matched to eligible resources, and energy efficiency is calculated. By comparing and collecting other relevant parameters, and substituting these parameters into a resource demand prediction model, the model outputs the resource score prediction value of the computing power resources to be allocated. Through detailed analysis of the utilization rate and latency information of heterogeneous computing power resources, the model can efficiently match the resources most suitable for the target processing task. By eliminating computing power resources with low latency scores, the overall resource utilization efficiency is improved, and resource waste is avoided. Combining multiple parameters (such as model, energy efficiency ratio, utilization rate, number of load tasks, and energy consumption ratio) for resource demand prediction allows for a more accurate assessment of the performance and value of the computing power resources to be allocated. This achieves effective management and optimized utilization of heterogeneous computing power resources.

[0033] The dynamic allocation management unit dynamically allocates and manages heterogeneous computing resources. Specifically, it sorts the heterogeneous computing resources from largest to smallest according to their pre-defined resource allocation values. If the number of target processing tasks is less than or equal to the number of heterogeneous computing resources, it selects a preset number of heterogeneous computing resources sequentially from the top of the list. Then, it distributes the target processing tasks equally among the edge nodes corresponding to the heterogeneous computing resources, where they are processed by the corresponding heterogeneous computing resources within the edge nodes. If the number of target processing tasks is greater than the number of heterogeneous computing resources, it sends the target processing tasks to the edge node corresponding to the top-ranked heterogeneous computing resource until its utilization rate equals the maximum utilization rate threshold, at which point it stops sending tasks. Then, it sends tasks to the edge node corresponding to the second-ranked heterogeneous computing resource, and so on, until all target processing tasks are distributed.

[0034] The fault detection and processing unit performs fault detection and analysis on heterogeneous computing resources. The specific process is as follows:

[0035] Acquire all temperatures corresponding to heterogeneous computing resources, compare each temperature with its corresponding preset maximum temperature, and count the number of temperatures equal to or higher than the maximum temperature to obtain the heterogeneous temperature count. Calculate the average temperature by averaging all temperatures. When the average temperature exceeds a set average threshold, acquire the voltage Uj and current Ij of the heterogeneous computing resources, j = 1, 2, ..., m, where m represents the acquisition time number and m represents the maximum value of that number. Substitute these values ​​into the formula... Output voltage fluctuation value SU and current fluctuation value SI; normalize the number of temperature variations, voltage fluctuation value, average temperature value, and current fluctuation value, and take the values ​​of the four, then substitute them into the formula. Obtain the fault detection value GYi of heterogeneous computing power resources; where β1, β2, β3 and β4 are the weighting factors corresponding to the number of abnormal temperatures YW, the average temperature value PJ, the voltage fluctuation value SU and the current fluctuation value SI, respectively.

[0036] Define two intervals: Interval 1 and Interval 2. When the fault detection value falls within Interval 1, shut down the edge node corresponding to the heterogeneous computing resource. Then, convert the fault detection value into a shutdown duration according to a certain ratio, execute the shutdown duration, and then restart. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. When the fault detection value falls within Interval 2, execute a self-check and repair program on the edge node corresponding to the heterogeneous computing resource. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. After the self-check and repair program is completed, recalculate the fault detection value of the heterogeneous computing resource. If the fault detection value still falls within Interval 2, mark the edge node corresponding to the heterogeneous computing resource as a fault node. Send the coordinates of the fault node, the name of the heterogeneous computing resource, and the fault detection value to the smart terminal of the corresponding maintenance personnel. The maintenance personnel will then go to the location of the fault node to repair or replace the heterogeneous computing resource.

[0037] By monitoring key parameters such as temperature, voltage, and current of heterogeneous computing resources in real time, potential faults can be detected and addressed promptly, improving system stability and reliability. Different handling measures are automatically implemented based on different fault detection ranges, achieving intelligent and automated fault handling. This reduces service interruptions and data loss caused by faults, improving system availability.

[0038] Example 2:

[0039] Based on Example 1, the dynamic allocation management unit also sets up a task queuing program. The task queuing program queues the target processing tasks in order according to the sorting rules, so as to ensure that the tasks are executed in order and reduce waiting time. The sorting rules are as follows: Select target processing tasks within a preset time period, which can be one minute, five minutes, ten minutes, or one hour, or other time periods. Identify urgency indicators for the target processing tasks. If an urgency indicator is detected, the target processing task is ranked first. If multiple target processing tasks with urgency indicators are detected, they are sorted according to the order in which they were received. Then, sort other target processing tasks without urgency indicators. Match the keywords of the target processing tasks with a set of preset keywords. Each preset keyword has a priority score. If the keywords of the target processing task match the preset keywords, the corresponding priority score is matched. Sum all the priority scores matched by the target processing task to obtain a total priority score. Sort according to the total priority score. If the total priority scores are the same, sort according to the order of time. After sorting, sort the target processing tasks that did not match a priority score according to the order of time. After sorting, the sorting rules are applied to the next preset time period.

[0040] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic allocation and management system for real-time monitoring of computing resources, comprising a central cloud and several edge nodes connected to the central cloud, wherein the central cloud is equipped with a real-time monitoring and prediction unit and a dynamic allocation and management unit, characterized in that, The real-time monitoring and prediction unit is used to collect and analyze the status information of the heterogeneous computing resources corresponding to the edge nodes in real time using monitoring tools and algorithms. The specific analysis process is as follows: The system acquires the target processing task, matches it with the required heterogeneous computing resources, and determines the utilization rate of these resources. If the utilization rate is less than a set threshold, the resource is marked as a pending allocation resource and assigned the number i. The system then acquires the status information of the pending allocation resources, analyzes all latency durations within a preset time period to obtain the latency score, and acquires the corresponding latency threshold. The system compares the latency score with the threshold; if the latency score is less than or equal to the threshold, the resource is removed; otherwise, it is discarded. The model of the computing power resource to be allocated is matched with several preset models, each of which corresponds to a model value. If the model of the computing power resource to be allocated matches a corresponding preset model, the model value of the matched preset model is marked as the model value XHi of the computing power resource to be allocated. The energy efficiency ratio NXI is obtained by dividing the number of floating-point operations of the computing power resource to be allocated by the power value. The utilization rate, number of load tasks, and energy consumption ratio of the computing power resource to be allocated are marked as LYi, FZi, and NXI, respectively. The values ​​are then substituted into the resource demand prediction model to output the resource allocation prediction value of the computing power resource to be allocated. Send the pre-set value of the computing power resources to be allocated to the dynamic allocation management unit; The dynamic allocation management unit is used to dynamically allocate and manage heterogeneous computing resources.

2. The dynamic allocation and management system for real-time monitoring computing resources according to claim 1, characterized in that, Heterogeneous computing resources include CPUs, GPUs, FPGAs, NPUs, and ASICs; status information includes performance data, utilization, number of workloads, and energy consumption data; performance data includes floating-point operations, latency, model, and temperature; energy consumption data includes power and energy efficiency ratio.

3. The dynamic allocation management system for real-time monitoring computing resources according to claim 1, characterized in that, The specific process for obtaining the deferred value of the computing power resources to be allocated is as follows: The delay durations are sorted according to the order of data collection. Adjacent delay durations are compared; if the delay duration collected earlier is greater than the delay duration collected later, the difference between the two is calculated to obtain the delay drop value. All delay drop values ​​are summed to obtain the total delay value. The average delay value is calculated by taking the average of all delay duration values. The maximum and minimum delay values ​​are then identified and labeled as the maximum and minimum delay durations, respectively. The total delay value, average delay value, maximum delay value, and minimum delay value are normalized and labeled as TZ, TP, TMAX, and TMIN, respectively. These values ​​are then substituted into the preset delay analysis model. Output the delay score TYi of the computing power resources to be allocated, where f1, f2, f3 and f4 are the weighting factors corresponding to the total delay, average delay, maximum delay and minimum delay, respectively.

4. The dynamic allocation and management system for real-time monitoring computing resources according to claim 1, characterized in that, The resource demand forecasting model is specifically as follows: Output the pre-allocated computing power resources ZFi; where hy1, hy2, and hy3 are model thresholds; For extended weighting factors, Energy efficiency ratio weighting factor; Utilization rate weighting factor; Energy consumption ratio weighting factor; The weighting factor for the number of load tasks is i = 1, 2, ..., n, where n is the maximum value of the number.

5. A dynamic allocation and management system for real-time monitoring computing resources according to claim 1, characterized in that, The specific process of dynamic allocation management is as follows: The heterogeneous computing resources are sorted from largest to smallest according to their pre-defined values. If the number of target processing tasks is less than or equal to the number of heterogeneous computing resources, a preset number of heterogeneous computing resources are selected sequentially from front to back according to the sorting. The target processing tasks are then sent to the edge nodes corresponding to the heterogeneous computing resources in sequence, and processed by the corresponding heterogeneous computing resources within the edge nodes. If the number of target processing tasks is greater than the number of heterogeneous computing resources, the target processing tasks are sent to the edge node corresponding to the top-ranked heterogeneous computing resource until its utilization rate equals the maximum utilization rate threshold, at which point the sending stops; then the tasks are sent to the edge node corresponding to the second-ranked heterogeneous computing resource; until all target processing tasks are distributed.

6. The dynamic allocation management system for real-time monitoring computing resources according to claim 1, characterized in that, The central cloud is also equipped with a fault detection and processing unit; the fault detection and processing unit is used to perform fault detection and analysis on heterogeneous computing resources, and the specific process is as follows: Acquire all temperatures corresponding to heterogeneous computing resources, compare each temperature with its corresponding preset maximum temperature, and count the number of temperatures equal to or higher than the maximum temperature to obtain the heterogeneous temperature count. Calculate the average temperature by averaging all temperatures. When the average temperature exceeds a set average threshold, acquire the voltage Uj and current Ij of the heterogeneous computing resources, j = 1, 2, ..., m, where m represents the acquisition time number and m represents the maximum value of that number. Substitute these values ​​into the formula... Output voltage fluctuation value SU and current fluctuation value SI; normalize the number of temperature variations, voltage fluctuation value, average temperature value, and current fluctuation value, and take the values ​​of the four, then substitute them into the formula. Obtain the fault detection value GYi of heterogeneous computing power resources; where β1, β2, β3 and β4 are the weighting factors corresponding to the number of abnormal temperatures YW, the average temperature value PJ, the voltage fluctuation value SU and the current fluctuation value SI, respectively. Define two intervals: Interval 1 and Interval 2. When the fault detection value falls within Interval 1, shut down the edge node corresponding to the heterogeneous computing resource. Then, convert the fault detection value into a shutdown duration according to a certain ratio, execute the shutdown duration, and then restart. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. When the fault detection value falls within Interval 2, execute a self-check and repair program on the edge node corresponding to the heterogeneous computing resource. Simultaneously, any unexecuted load tasks on the edge node are fed back to the dynamic allocation management unit for reallocation. After the self-check and repair program is completed, recalculate the fault detection value of the heterogeneous computing resource. If the fault detection value still falls within Interval 2, mark the edge node corresponding to the heterogeneous computing resource as a fault node. Send the coordinates of the fault node, the name of the heterogeneous computing resource, and the fault detection value to the smart terminal of the corresponding maintenance personnel. The maintenance personnel will then go to the location of the fault node to repair or replace the heterogeneous computing resource.

Images