Integrated management system based on artificial intelligence computing platform

Abstract

The invention discloses an integrated management system based on an artificial intelligence computing platform, and relates to the technical field of integrated management, and the system comprises the following steps: monitoring and obtaining resources, and obtaining heterogeneous computing resource state information; performing resource health assessment according to the heterogeneous computing resource state information to obtain a resource availability index; performing task resource matching according to the resource availability index to obtain a resource allocation scheme; performing calculation task scheduling according to the resource allocation scheme to obtain a task execution sequence; performing task execution monitoring according to the task execution sequence to obtain a task real-time execution state; and performing resource adjustment according to the task real-time execution state to obtain a resource allocation adjustment result. According to the method, the matching accuracy and the resource utilization rate are improved, resource waste caused by single resource bottleneck is avoided, and the load balance and the overall operation stability of each node are improved.

Description

Technical Field

[0001] The present invention relates to the field of integrated management technology, and in particular to an integrated management system based on an artificial intelligence computing platform. Background Art

[0002] With the rapid development of artificial intelligence computing platforms and the increasingly diversified and dynamic demand for computing resources from massive tasks, how to control the status of computing resources, efficiently utilize them, and reasonably allocate them to achieve integrated management has become a key technical issue.

[0003] In the existing technology, there are deficiencies in task resource allocation: In the existing technology, task resource matching mostly adopts fixed heuristic algorithms or simple resource satisfaction judgments, such as only judging whether the resources required for the task are less than the current available resources of the node, and lacks integration with real-time resource availability indicators. It is difficult to reflect the actual demand and matching degree of the task for resources, affecting the stability of task execution and the overall load balance of the system. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an integrated management system based on an artificial intelligence computing platform to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides an integrated management system based on an artificial intelligence computing platform, comprising the following steps: S1, resource monitoring and acquisition, obtaining heterogeneous computing resource status information; S2. Perform resource health assessment based on heterogeneous computing resource status information to obtain resource availability indicators; S3. Match task resources based on resource availability indicators to obtain a resource allocation plan; S4. Schedule computing tasks according to the resource allocation plan to obtain a task execution sequence; S5. Monitor task execution according to the task execution sequence to obtain the real-time execution status of the task; S6. Adjust resources according to the real-time execution status of the task to obtain a resource allocation adjustment result.

[0006] To further optimize this technical solution, the resource health assessment in step S2 includes: Based on the obtained real-time status information of heterogeneous computing resources, a resource health assessment model is used to perform a nonlinear assessment of heterogeneous computing resources by integrating the status and historical stability of each resource to obtain the resource availability index of each computing node.

[0007] To further optimize this technical solution, the resource health assessment model includes:

[0008] in: : resource availability indicator of the jth node; : Sigmoid function; : Weighting coefficient of current status score; : weighting coefficient of historical stability score; : weighting coefficient of coupling term; : The current status score of the jth node; : The historical stability score of the j-th node.

[0009] To further optimize this technical solution, the current status score includes:

[0010] in: : CPU idle rate of the jth node; : GPU idle rate of the jth node; : The memory idle rate of the jth node; : Weight coefficient of CPU idle rate; : Weight coefficient of GPU idle rate; : Weight coefficient of memory idle rate; Calculate the node's current status score based on the current CPU and GPU idle rates and memory idle rates.

[0011] To further optimize this technical solution, the historical stability score includes:

[0012] in: : the number of historical observation periods; : The abnormal rate of node j in the kth historical observation period; The historical stability score is obtained by calculating the normality rate of nodes in all historical observation periods.

[0013] To further optimize this technical solution, the task resource matching in step S3 includes: Based on the obtained resource availability indicators and the resource requirements of the tasks to be executed, the task matching model is used to calculate the matching scores between tasks and resource nodes, and tasks are allocated based on the matching scores to obtain a resource allocation plan.

[0014] To further optimize this technical solution, the task matching model includes:

[0015] in: : The matching score between task i and node j; : The matching degree between task i and the resource requirements of the jth node; : historical adaptation compatibility factor between task i and node j; : Weight coefficient of resource demand matching; : Weight coefficient of resource availability indicator; : Weight coefficient of historical adaptation compatibility factor.

[0016] To further optimize this technical solution, the resource demand matching degree includes:

[0017] in: : Estimated CPU usage of task i; : Estimated GPU usage of task i; : Estimated memory usage of task i; : The weight coefficient of the remaining available CPU; : Weight coefficient of the remaining available GPUs; : The weight coefficient of the remaining available memory; The resource requirement matching degree between the node and the task is calculated based on the available resources of the node and the required resources of the task.

[0018] To further optimize this technical solution, the historical adaptation compatibility factors include:

[0019] in: : The number of successful executions of task i on node j; : The total number of times task i is scheduled on node j; : A small constant to prevent division by zero; The historical adaptation compatibility factor between the task and the node is calculated based on the ratio of the number of successful executions of the task on the node to the total number of scheduling times.

[0020] To further optimize this technical solution, the functional modules include: Resource status acquisition module, resource health assessment module, task resource matching module, resource allocation execution module, task operation monitoring module, and resource dynamic adjustment module.

[0021] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of an integrated management system based on an artificial intelligence computing platform as described in the first aspect of the present invention are implemented.

[0022] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of an integrated management system based on an artificial intelligence computing platform as described in the first aspect of the present invention are implemented.

[0023] Compared with the existing technology, the present invention provides an integrated management system based on an artificial intelligence computing platform, which has the following beneficial effects: This integrated management system based on an artificial intelligence computing platform takes into account multiple resource dimensions through a task matching model, improves matching accuracy and resource utilization, avoids resource waste caused by a single resource bottleneck, and responds to node resource changes in real time. It matches the historical task execution status of the node, thereby improving the load balance of each node and the overall operation stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 This is a flow chart of an integrated management system based on an artificial intelligence computing platform proposed by the present invention; Figure 2 This is a flow chart of a resource health assessment model for an integrated management system based on an artificial intelligence computing platform proposed in the present invention; Figure 3 This is a flow chart of a task matching model for an integrated management system based on an artificial intelligence computing platform proposed by the present invention; Figure 4 This is a flow chart of an execution status scoring model for an integrated management system based on an artificial intelligence computing platform proposed by the present invention; Figure 5 This is a module diagram of an integrated management system based on an artificial intelligence computing platform proposed by the present invention. DETAILED DESCRIPTION

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0028] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it designate a separate or selective embodiment that is mutually exclusive with other embodiments.

[0029] Example 1:

[0030] Reference Figures 1 to 4 , which is the first embodiment of the present invention, provides an integrated management system based on an artificial intelligence computing platform, comprising the following steps: S1. Resource monitoring and acquisition to obtain heterogeneous computing resource status information.

[0031] In this embodiment, the resource monitoring and acquisition includes: In integrated management, computing resources are diverse and heterogeneous. Common resources include CPUs, GPUs, TPUs, and other computing units. Each resource type has different performance, architecture, and load conditions. Since the efficiency of task scheduling, resource allocation, and management depends on accurate understanding of the status of each computing resource, real-time understanding of the operating status of computing resources is a prerequisite for integrated management. The lack of real-time, accurate resource status information will lead to resource scheduling decisions based on incomplete or outdated data, increasing task execution failure rates, reducing resource utilization efficiency, and even causing system overload or failure risks. Therefore, comprehensive, real-time monitoring and acquisition of computing resources are necessary.

[0032] The main purpose of this step is to monitor and obtain the status information of heterogeneous computing resources, including core indicators such as the load, memory usage, temperature, and power consumption of each computing node, so as to accurately reflect the resource status, understand the current status of resources, support the implementation of subsequent steps, promptly detect hardware anomalies, prevent overload and damage, ensure the stable operation of the computing platform, assist in identifying idle or inefficiently used resources, promote the rational allocation and scheduling of resources, and improve resource utilization efficiency.

[0033] This step installs a monitoring and collection component on each heterogeneous computing node, regularly accesses the node's performance counters and sensor data, and continuously collects status data of local hardware resources, including core indicators such as CPU usage and idle rate, GPU usage and idle rate, memory usage, device temperature and power consumption. The collected raw indicators are basically cleaned and formatted, including removing abnormal values, converting units, and unifying timestamps to ensure unified standards and comparability of data across nodes. In addition, to ensure the consistency of status data in time and content in heterogeneous environments, a time synchronization mechanism is adopted to ensure that all collected data is accompanied by a standard and unified timestamp, thereby improving the accuracy and timeliness of the monitoring system.

[0034] S2. Perform resource health assessment based on heterogeneous computing resource status information to obtain resource availability indicators.

[0035] In this embodiment, the resource health assessment includes: Based on the obtained real-time status information of heterogeneous computing resources, a resource health assessment model is used to perform nonlinear assessment of heterogeneous computing resources by integrating the status and historical stability of each resource. This achieves a unified and standardized resource health assessment, thereby obtaining the resource availability index of each computing node, providing a basic basis for subsequent resource allocation and task scheduling.

[0036] Furthermore, the resource health assessment model includes:

[0037] in: : The resource availability index of the jth node, ranging from 0 to 1; : Sigmoid function (existing technology), used to normalize the output; : The weighting coefficient of the current status score, ranging from 0 to 1, and the sum of the weighting coefficients is 1; : weighting coefficient of historical stability score, ranging from 0 to 1; : weighting coefficient of the coupling term, ranging from 0 to 1; : The current status score of the jth node, ranging from 0 to 1. The larger the score, the more available resources the node has. : The historical stability score of the jth node, ranging from 0 to 1, where a larger value represents a more stable historical node.

[0038] Furthermore, the current status score includes:

[0039] in: : The CPU idle rate of the jth node, ranging from 0 to 1, is calculated by the ratio of the number of unused CPU cores to the total number of CPU cores; : The GPU idle rate of the jth node, ranging from 0 to 1, is calculated by the ratio of the number of unused GPU cores to the total number of GPU cores; : The memory idle rate of the jth node, ranging from 0 to 1, is calculated by the ratio of the unused memory size to the total memory size; : The weight coefficient of the CPU idle rate ranges from 0 to 1, and the sum of the weight coefficients is 1; : GPU idle rate weight coefficient, ranging from 0 to 1; : The weight coefficient of memory idle rate, ranging from 0 to 1; Calculate the node's current status score based on the current CPU and GPU idle rates and memory idle rates.

[0040] Furthermore, the historical stability score includes:

[0041] in: : The number of historical observation periods, for example, the number of hourly samples within 24 hours; : The abnormal rate of node j in the kth historical observation period, that is, the proportion of abnormal time in this period, ranging from 0 to 1. Abnormalities include downtime and disconnection; The historical stability score is obtained by calculating the normality rate of nodes in all historical observation periods.

[0042] This model describes how to obtain the resource availability indicators of computing nodes based on the real-time status information and historical data of heterogeneous computing resources.

[0043] Traditional resource health assessment methods usually process various indicators in a linear manner such as weighted average, which makes it difficult to reflect the nonlinear coupling characteristics of actual load risks, resulting in insufficient rationality of the obtained resource availability indicators and easily causing node overload during scheduling. However, this model integrates the status and historical stability of various resources, which can better reflect the comprehensive health level of the node, improve the rationality of the resource availability indicators, and enhance the stability of the node during scheduling.

[0044] The steps for using this model include: Data acquisition: Obtain the usage of each node resource from step S1, including CPU idle rate , GPU idle rate , memory idle rate , and obtain the abnormal rate of each observation period node from historical data ; Score calculation: Calculate the current status score based on the obtained data and historical stability scores ; Resource health assessment: The resource health is assessed based on the calculated score to obtain the node resource availability index of each node. This index reflects the overall health level of the node and serves as a key input for subsequent resource scheduling and task allocation, ensuring that tasks are allocated to nodes with abundant and stable available resources.

[0045] S3. Match task resources according to resource availability indicators to obtain a resource allocation plan.

[0046] In this embodiment, the task resource matching includes: This step uses the task matching model to calculate the matching score between tasks and resource nodes based on the obtained resource availability indicators and the resource requirements of the tasks to be executed. Tasks are then allocated based on the matching scores to achieve the optimal task allocation and obtain a resource allocation plan.

[0047] Furthermore, the task matching model includes:

[0048] in: : The matching fitness score between task i and node j, ranging from 0 to 1. The higher the score, the higher the fitness between task and node, and the more likely the task will be assigned to this node. : The matching degree of the resource requirements of task i and node j, usually ranging from 0 to 1. A larger value indicates that the idle resources of the node are greater than the resources required by the task. When it is equal to 0, it means that the estimated resource usage of the task is equal to the idle resources of the node. When it is less than 0, it means that any of the estimated resource usage of the task is greater than the idle resources of the node, and the task cannot be executed on this node. : The historical adaptation compatibility factor between task i and node j, reflecting the historical execution success rate of the task on this node, ranging from 0 to 1; : The weight coefficient of resource demand matching, ranging from 0 to 1, and the sum of the weight coefficients is 1; : The weight coefficient of resource availability indicator, ranging from 0 to 1; : The weight coefficient of the historical adaptation compatibility factor, ranging from 0 to 1.

[0049] Furthermore, the resource demand matching degree includes:

[0050] in: : The expected CPU usage of task i, that is, the minimum CPU usage; : The expected GPU usage of task i, that is, the minimum GPU usage; : The expected memory usage of task i, that is, the minimum memory usage; : The weight coefficient of the remaining available CPU, ranging from 0 to 1, and the sum of the weight coefficients is 1; : The weight coefficient of the remaining available GPUs, ranging from 0 to 1; : The weight coefficient of the remaining available memory, ranging from 0 to 1; The resource requirement matching degree between the node and the task is calculated based on the available resources of the node and the required resources of the task.

[0051] Furthermore, the historical adaptation compatibility factors include:

[0052] in: : The number of successful executions of task i on node j; : The total number of times task i is scheduled on node j; : A small constant to prevent division by zero; The historical adaptation compatibility factor between the task and the node is calculated based on the ratio of the number of successful executions of the task on the node to the total number of scheduling times.

[0053] The model describes how to match tasks with nodes based on the comparison of the resources required by the task with the existing resources of the node and the historical task success rate.

[0054] Traditional task resource matching methods usually only use fixed rules or thresholds to match a single resource dimension, ignoring historical adaptation compatibility, which can easily lead to problems such as uneven resource utilization, node overload, and high task failure rate. However, this model comprehensively considers multiple resource dimensions when matching tasks and nodes, improving matching accuracy and resource utilization, avoiding resource waste caused by single resource bottlenecks, and responding to node resource changes in real time. It matches based on the node's historical task execution status, improving the load balancing of each node and the overall operation stability.

[0055] The steps for using the above model include: Task requirement acquisition: obtain the resource requirements of the task to be executed and the available resources of each node, including the estimated CPU usage , estimated GPU usage , estimated memory usage , CPU idle rate , GPU idle rate , memory idle rate ; Node matching: Based on the resource requirements of the task to be executed and the idle resources of the node, the matching degree between the task and the node's resource requirements is calculated. and historical adaptation compatibility factor , and when any resource requirement of the task to be executed is greater than the idle resources of the node, the node cannot be scheduled to execute the task. and historical adaptation compatibility factor , combined with resource availability indicators , calculate the matching adaptation score ; Allocation plan acquisition: based on the calculated matching adaptation score , nodes are allocated for each task to be executed. The larger the matching adaptation score, the more suitable the node is for executing the task. Under the premise of ensuring that the idle resources of each node are greater than the resource requirements of the assigned task, the final resource allocation plan is obtained.

[0056] S4. Schedule computing tasks according to the resource allocation plan to obtain a task execution sequence.

[0057] In this embodiment, the computing task scheduling includes: The resource allocation scheme is obtained from step S4, which solves the problem of task allocation on the node. In actual execution, each node may undertake multiple tasks. If scheduling and sorting are not performed, resource conflicts or disordered task queuing will occur, affecting the overall computing efficiency. Therefore, it is also necessary to reasonably arrange the execution order and timing of each task on the node according to the characteristics and resource consumption of each task to obtain the task execution sequence. Under the premise of ensuring that the resources of each node do not conflict, the waiting time of the task and the idle time of the node are reduced as much as possible, the number of tasks completed per unit time is increased, the resource preemption and queuing conflicts in task execution are reduced, and the system stability and resource utilization are improved.

[0058] This step calculates the sorting weights for tasks on each node based on the resource allocation plan. The sorting weights are calculated based on the expected task runtime (estimated based on historical data), the task startup delay tolerance (determined by whether it is a low-latency task or a batch task), and the task resource consumption (obtained by combining CPU, GPU, and memory resource consumption). A smaller value indicates a higher scheduling priority. Tasks on each node are sorted according to their scheduling priority to form the final task execution sequence for the node. This avoids execution blockage and resource idleness caused by resource competition, thereby improving overall throughput and computing efficiency.

[0059] S5. Monitor the task execution according to the task execution sequence to obtain the real-time execution status of the task.

[0060] In this embodiment, the task execution monitoring includes: Based on the obtained task execution sequence, the actual execution status of the task on the node is dynamically monitored, and the execution of the task is quantitatively evaluated using the execution status scoring model, thereby accurately reflecting the resource utilization efficiency and progress of task execution and supporting dynamic adjustment of resource allocation.

[0061] Furthermore, the execution status scoring model includes:

[0062] in: : The execution status score of task i on node j at time t, which reflects the health of the current task execution. The closer the value is to 1, the healthier and more stable the current task execution is. : The actual CPU usage of task i on node j at time t is very close to the expected CPU usage, which indicates that resource allocation is accurate and task execution is normal. A CPU usage lower than the expected value indicates insufficient resource supply and restricted task execution. A CPU usage higher than the expected value indicates that resource allocation exceeds task demand, which may lead to reduced resource utilization efficiency. : The actual GPU usage of task i on node j at time t is compared with the expected GPU usage in the same way as the CPU usage above; : The actual memory usage of task i on node j at time t is compared with the expected memory usage in the same way as the CPU usage above; : The execution progress of task i on node j at time t, ranging from 0 to 1, where 0 represents just executed and 1 represents completed; , , , : Weight coefficient, ranging from 0 to 1, and the sum of weight coefficients is 1.

[0063] This model describes how to quantitatively evaluate the execution of tasks based on the execution status of tasks on nodes.

[0064] Traditional task execution monitoring relies heavily on a single resource utilization metric (such as CPU utilization), lacks a comprehensive evaluation combined with task demand matching, cannot accurately reflect the true operational health of the task, and is difficult to assign reasonable weights to different resource types. This model integrates matching adaptation scores with real-time usage data, making up for the blind spots of single resource monitoring and improving monitoring accuracy. It also uses multi-dimensional resource utilization and task progress factor weighting to adapt to the differences in sensitivity of diverse heterogeneous tasks to different resources, thereby improving model adaptability.

[0065] The steps for using the model include: Data acquisition: Based on the dynamic monitoring data of the actual execution status of the task on the node, the actual CPU usage of the task on the node is obtained. , actual GPU usage , actual memory usage and execution progress , and obtain the matching score between the task and the node from step S3 ; Execution status score calculation: Based on the obtained data, calculate the execution status score of the task on the node ,This score reflects the health of the current task execution; Execution status analysis: Based on the calculated execution status score , the analysis obtains the preliminary execution status. The closer the score is to 1, the more accurate the resource scheduling, the higher the task utilization efficiency, and the healthier and more stable the execution status. A score greater than 1 indicates that there is resource pressure and it is necessary to monitor the task operation trend and make dynamic adjustments. A score less than 1 indicates that the resource usage is less than the demand, and there is a possibility of insufficient resource utilization or inadequate task execution. The more specific execution status is analyzed based on the actual GPU, CPU and memory usage. For example, if the CPU and memory usage are very close to the demand, while the GPU is slightly greater than the demand, it means that the task is in a relatively healthy running state, but the GPU usage is slightly exceeded, which deserves attention.

[0066] S6. Adjust resources according to the real-time execution status of the task to obtain a resource allocation adjustment result.

[0067] In this embodiment, the resource adjustment includes: In artificial intelligence computing platforms, the execution environment and resource status of computing tasks are highly dynamic. As the task progresses, resource requirements and usage continue to change, and a single resource matching solution is difficult to ensure that the task is always in the optimal execution state. In addition, there are problems such as resource competition and node load fluctuations between tasks. If resource allocation is not adjusted in real time, it may lead to resource waste or task performance degradation. Therefore, it is necessary to dynamically adjust resource allocation based on the real-time execution status of the task obtained from step S5, so as to improve resource utilization, reasonably increase or decrease resource quotas, ensure smooth task execution, minimize performance degradation or task blocking caused by insufficient resources, and improve load balancing between nodes.

[0068] Based on the obtained real-time execution status of the task, this step introduces a resource adjustment prediction mechanism based on the task's historical operation trajectory. By storing and analyzing multiple rounds of historical task operation data, it predicts the changing trends of indicator requirements, anticipates future resource needs, implements pre-adjustments, and reduces the allocation delay of subsequent resource adjustments. Based on the detection results, the resources used by the task in real-time execution are compared with its allocated resources to determine whether there is a resource allocation mismatch (judgment criteria include the task's actual resource utilization rate being significantly lower than the allocated resources, the task's progress being blocked or slowed, etc.). Taking into account the task priority, resource availability, and overall platform load, a resource adjustment plan is formulated (the plan may include increasing CPU cores, GPU units, or memory capacity, or reclaiming over-allocated resources, etc.) to ensure that the adjustment process does not affect the current stable operation of the task. Resource adjustments are made according to the resource adjustment plan, and the adjustment results are monitored in real time to confirm whether the resource adjustment achieves the expected goals. If not, further adjustments and optimizations are made, thereby achieving dynamic and refined management of task resource allocation and adapting to task load fluctuations and changes in the status of the AI ​​computing platform.

[0069] Example 2:

[0070] Reference Figure 5 , which is the second embodiment of the present invention, provides an integrated management system based on an artificial intelligence computing platform, including the following functional modules: Resource status acquisition module: collects resource status information of each node in the computing platform, including the number of CPU cores, number of GPU units, memory capacity and usage, to form complete resource basic data, providing a basis for subsequent resource health assessment; Resource health assessment module: Based on the collected resource status information, it calculates the resource availability index of each node to reflect the health status and actual available capacity of the node resources; Task resource matching module: combines task requirements and resource health assessment results to calculate the matching degree between tasks and nodes and determine the optimal resource allocation plan; Resource allocation execution module: allocates resources to each computing node based on the task resource matching results, and implements specific resource allocation and scheduling; Task operation monitoring module: monitors the task operation status in real time and analyzes the task execution status and resource usage status based on resource allocation results; Resource dynamic adjustment module: adjusts resources according to the real-time execution status of tasks, realizes dynamic optimization of resource allocation, and ensures system load balancing and task performance.

[0071] Example 3:

[0072] This embodiment also provides a computer device suitable for an integrated management system based on an artificial intelligence computing platform, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement an integrated management system based on an artificial intelligence computing platform as proposed in the above embodiment.

[0073] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, an integrated management system based on an artificial intelligence computing platform as proposed in the above embodiment is implemented.

[0074] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse.

[0075] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0076] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0077] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0078] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0079] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An integrated management system based on an artificial intelligence computing platform, characterized in that: The following steps are involved: S1, resource monitoring and acquisition, obtaining heterogeneous computing resource status information; S2. Perform resource health assessment based on heterogeneous computing resource status information to obtain resource availability indicators; S3. Match task resources based on resource availability indicators to obtain a resource allocation plan; S4. Schedule computing tasks according to the resource allocation plan to obtain a task execution sequence; S5. Monitor task execution according to the task execution sequence to obtain the real-time execution status of the task; S6. Adjust resources according to the real-time execution status of the task to obtain a resource allocation adjustment result.

2. The integrated management system based on an artificial intelligence computing platform according to claim 1, characterized in that: The resource health assessment in step S2 includes: Based on the obtained real-time status information of heterogeneous computing resources, a resource health assessment model is used to perform a nonlinear assessment of heterogeneous computing resources by integrating the status and historical stability of each resource to obtain the resource availability index of each computing node.

3. The integrated management system based on an artificial intelligence computing platform according to claim 2, characterized in that: The resource health assessment model includes: ， in: : resource availability indicator of the jth node; : Sigmoid function; : Weighting coefficient of current status score; : weighting coefficient of historical stability score; : weighting coefficient of coupling term; : The current status score of the jth node; : The historical stability score of the j-th node.

4. The integrated management system based on an artificial intelligence computing platform according to claim 3 is characterized in that: The current status score includes: ， in: : CPU idle rate of the jth node; : GPU idle rate of the jth node; : The memory idle rate of the jth node; : Weight coefficient of CPU idle rate; : Weight coefficient of GPU idle rate; : Weight coefficient of memory idle rate; Calculate the node's current status score based on the current CPU and GPU idle rates and memory idle rates.

5. The integrated management system based on an artificial intelligence computing platform according to claim 3 is characterized in that: The historical stability score includes: ， in: : the number of historical observation periods; : The abnormal rate of node j in the kth historical observation period; The historical stability score is obtained by calculating the normality rate of nodes in all historical observation periods.

6. The integrated management system based on an artificial intelligence computing platform according to claim 1, characterized in that: The task resource matching in step S3 includes: Based on the obtained resource availability indicators and the resource requirements of the tasks to be executed, the task matching model is used to calculate the matching scores between tasks and resource nodes, and tasks are allocated based on the matching scores to obtain a resource allocation plan.

7. The integrated management system based on an artificial intelligence computing platform according to claim 6, characterized in that: The task matching model includes: ， in: : The matching score between task i and node j; : The matching degree between task i and the resource requirements of the jth node; : historical adaptation compatibility factor between task i and node j; : Weight coefficient of resource demand matching; : Weight coefficient of resource availability indicator; : Weight coefficient of historical adaptation compatibility factor.

8. The integrated management system based on an artificial intelligence computing platform according to claim 7, characterized in that: The resource demand matching degree includes: ， in: : Estimated CPU usage of task i; : Estimated GPU usage of task i; : Estimated memory usage of task i; : The weight coefficient of the remaining available CPU; : Weight coefficient of the remaining available GPUs; : The weight coefficient of the remaining available memory; The resource requirement matching degree between the node and the task is calculated based on the available resources of the node and the required resources of the task.

9. The integrated management system based on an artificial intelligence computing platform according to claim 7, characterized in that: The historical adaptation compatibility factors include: ， in: : The number of successful executions of task i on node j; : The total number of times task i is scheduled on node j; : A small constant to prevent division by zero; The historical adaptation compatibility factor between the task and the node is calculated based on the ratio of the number of successful executions of the task on the node to the total number of scheduling times.

10. The integrated management system based on an artificial intelligence computing platform according to claim 1, characterized in that: The functional modules include: Resource status acquisition module, resource health assessment module, task resource matching module, resource allocation execution module, task operation monitoring module, and resource dynamic adjustment module.

Images