AI-based computing power resource intelligent scheduling system and method

Abstract

The invention relates to the technical field of computing power management, and discloses an AI-based computing power resource intelligent scheduling system and method, and the system comprises a resource monitoring module, a multi-dimensional matching module and a computing power scheduling module which are in communication connection in sequence. The method corresponds to the system. According to the application, the resource monitoring module is utilized to provide a data basis by monitoring and predicting the use condition of the computing power resource, so that the system can timely deal with the dynamic change of the resource; the multi-dimensional matching module analyzes task multi-dimensional information and matches the most suitable computing center based on a computing power rule and a hardware rule, so that the defects of a traditional method in the aspect of task resource matching are effectively overcome; the computing power scheduling module updates actual scheduling according to a simulation result by means of simulation scheduling and an optimization mechanism, so that the scheduling accuracy and the resource utilization efficiency are greatly improved, the problems of imbalance of multi-task scheduling, insufficient system expansibility and the like are successfully solved, the stability and the overall efficiency of the system are remarkably enhanced, and the service life of the system is prolonged. And the improvement of the intelligence level of computing power scheduling is powerfully promoted.

Description

Technical Field

[0001] The present application relates to the field of computing power management technology, and specifically to an AI-based intelligent scheduling system and method for computing power resources. Background Art

[0002] With the rapid development of artificial intelligence technology, the demand for computing power has exploded, and the importance of computing power scheduling in distributed computing environments has become increasingly prominent. In the current technical landscape, although traditional computing power scheduling methods have a certain infrastructure, they still have many limitations, specifically manifested in: In terms of resource dynamics, scheduling decisions are mostly based on resource snapshots at the beginning of a task. In actual scenarios, computing center resources often change dramatically due to emergencies such as hardware failures and urgent insertion of new tasks. The lack of an effective response mechanism can easily lead to task execution interruptions or delays, seriously affecting system stability.

[0003] In terms of task resource matching, computing centers are mostly screened based on task type and simple resource quantity, without fully considering the strict requirements of special tasks for specific resource combinations and performance indicators (such as specific GPU architecture, memory bandwidth, etc.), making it difficult to achieve the optimal match between tasks and computing centers, reducing task execution efficiency.

[0004] When scheduling multiple tasks, simply scheduling by priority order can easily lead to an imbalance in resource allocation when facing a large number of complex tasks. Low-priority small tasks may wait for a long time, while high-priority tasks monopolize resources, resulting in idle resources and backlog of tasks, making it impossible to achieve efficient resource utilization and fast task turnover.

[0005] In addition, with the continuous increase in the number of computing centers and task types, the complexity of its hierarchical scheduling strategy has increased dramatically, the scheduling delay has increased significantly, it is difficult to adapt to large-scale distributed computing environments, and the system scalability faces severe challenges.

[0006] Chinese patent number CN116643893B discloses a scheduling method, scheduling device, storage medium and server for computing tasks, but the invention is insufficient in solving the above-mentioned problems.

[0007] In summary, a new technical solution for AI-based intelligent computing power scheduling is urgently needed to effectively solve the above problems and improve the intelligence level and overall efficiency of computing power scheduling. Summary of the invention

[0008] The purpose of this application is to provide an AI-based computing resource intelligent scheduling system and method to solve the technical problems raised in the above background technology.

[0009] To achieve the above objectives, this application discloses the following technical solutions: In a first aspect, the present application discloses an AI-based intelligent scheduling system for computing resources, including a resource monitoring module, a multi-dimensional matching module, and a computing resource scheduling module that are sequentially connected in communication; The resource monitoring module is configured to: monitor the real-time usage of computing resources and generate real-time resource data, and predict the usage of computing resources within a preset prediction period based on the multi-dimensional information of the tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information includes at least the computing power requirements and hardware requirements of the tasks; The multi-dimensional matching module is configured to: use the preset computing power rules and hardware rules, in combination with the multi-dimensional information, match the computing center corresponding to the multi-dimensional information, and obtain the resource allocation amount; wherein the computing power rules correspond to the computing power requirements and enable the matched computing center to meet the computing power requirements, and the hardware rules correspond to the hardware requirements and enable the matched computing center to meet the hardware requirements; The computing power scheduling module is configured as follows: performing a first computing power resource scheduling based on the matched computing center, performing a simulated computing power resource scheduling based on the resource prediction data, updating the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and executing after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform a simulated scheduling based on the resource prediction data to obtain a resource optimization amount.

[0010] Preferably, the computing power rule is specifically: Analyze the CPU computing power requirements, GPU computing power requirements, and memory requirements of each task; Obtain the available CPU computing power, available GPU computing power, available memory and computing power stability index of each computing center; wherein the computing power stability index is used to characterize the stable operation of the computing power of the computing center based on the time series; The computing power adaptation of each task assigned to each computing center is calculated based on the CPU computing power requirement, the GPU computing power requirement, the memory requirement, the available CPU computing power, the available GPU computing power, the available memory and the computing power stability index, and the computing center that meets the preset computing power adaptation threshold is defined as the preliminary matching result of the task and output; wherein the computing power adaptation is used to characterize the degree of adaptation of the computing power of the computing center when completing the task.

[0011] Preferably, the hardware rules are specifically: A hardware compatibility matrix is ​​constructed based on the hardware requirements of different task types and the hardware architecture of the computing center. The hardware compatibility matrix stores the compatibility degree between different task types and the hardware architecture. Analyze the task type of each task and match the corresponding compatibility level, obtain the hardware stability index of the computing center, and calculate the hardware adaptability based on the compatibility level and the hardware stability index; wherein the hardware stability index is used to characterize the stable operation of the hardware of the computing center based on the time series, and the hardware adaptability is used to characterize the adaptability of the hardware of the computing center when completing the task; The resource allocation amount is calculated based on the hardware adaptability and the computing power adaptability, and the preliminary matching results are screened based on the resource allocation amount, and the computing center corresponding to the maximum value of the resource allocation amount is determined as the matched computing center corresponding to the multi-dimensional information and output; wherein the resource allocation amount is used to characterize the computing power of the computing center and the degree of hardware adaptation when completing the task.

[0012] Preferably, the updating of the first computing resource scheduling based on the result of the simulated computing resource scheduling to obtain the second computing resource scheduling is specifically: Obtain the resource allocation amount of the computing center to which the task is allocated in the first computing power resource scheduling, and obtain the resource optimization amount obtained after the simulation scheduling; Based on the resource allocation amount and the resource optimization amount, combined with a preset update step coefficient, an update amount is calculated, and the resource allocation amount is adjusted based on the update amount to obtain a second computing power resource scheduling; wherein the update amount is used to optimize the first computing power resource scheduling.

[0013] Preferably, the verification is performed after the second computing power resource is scheduled, specifically: Obtain a first execution time corresponding to the scheduling of the first computing resource in the computing center, and obtain a second execution time corresponding to the scheduling of the second computing resource in the computing center; The task execution efficiency index is calculated based on the resource allocation amount, the resource optimization amount, the first execution time and the second execution time; when the task execution efficiency index is greater than the preset task execution efficiency index threshold, it is determined that the second computing power resource scheduling verification is passed, otherwise, continue to update the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain the second computing power resource scheduling; wherein, the task execution efficiency index is used to characterize the change in the task execution efficiency of the second computing power resource scheduling compared with the first computing power resource scheduling.

[0014] Preferably, when performing the first computing power resource scheduling, if there is a task with time requirements, the required completion time of the task is obtained; based on the current time and the estimated execution time of the task with time requirements, it is determined whether the required completion time is met, if so, no operation is performed, otherwise based on the required completion time, the resource allocation amount improvement coefficient is calculated based on the required completion time, the current time and the estimated execution time, and the update amount is recalculated based on the resource allocation amount improvement coefficient; wherein, the resource allocation amount improvement coefficient is used to optimize the update amount based on the time requirements of the task.

[0015] Preferably, the simulated computing power resource scheduling includes calculating the resource optimization amount obtained after the simulated scheduling; wherein the calculation process of the resource optimization amount is specifically as follows: Obtain the resource allocation amount of the computing center to which the task is assigned in the first computing power resource scheduling, combine the preset discount factor and task transfer probability, calculate the simulated resource allocation amount at each time in the prediction period, obtain the maximum value of the simulated resource allocation amount, define the maximum value as the resource optimization amount and output it.

[0016] Preferably, when performing the second computing power resource scheduling, if there is a large-scale parallel task, the task parallelism of the large-scale parallel task is obtained; the parallel processing capability of the computing center is obtained, the estimated parallel efficiency of the large-scale parallel task in the computing center is calculated, and the resource optimization amount is updated and output based on the estimated parallel efficiency; wherein the estimated parallel efficiency is used to characterize the processing capability of the computing center for large-scale parallel tasks.

[0017] Preferably, the system further comprises an adjustment feedback module which is in communication with the computing power scheduling module, and the adjustment feedback module is configured as follows: After each computing power scheduling is completed, the deviation between the actual data of task execution and the expected data is collected, and the parameter sets in the multidimensional matching module and the computing power scheduling module are updated based on the preset learning rate and maximum deviation value.

[0018] In a second aspect, the present application discloses an AI-based intelligent scheduling method for computing resources, which is applicable to the AI-based intelligent scheduling system for computing resources as described above, and the method includes: S1: monitor the real-time usage of computing resources and generate real-time resource data, and predict the usage of computing resources within a preset prediction period based on the multi-dimensional information of tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information at least includes the computing power requirements and hardware requirements of the tasks; S2: using preset computing power rules and hardware rules, combined with the multi-dimensional information, to match a computing center corresponding to the multi-dimensional information; wherein the computing power rules correspond to the computing power requirements and enable the matched computing center to meet the computing power requirements, and the hardware rules correspond to the hardware requirements and enable the matched computing center to meet the hardware requirements; S3: Perform a first computing power resource scheduling based on the matched computing center, perform a simulated computing power resource scheduling based on the resource prediction data, update the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and execute after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform simulated scheduling based on the resource prediction data.

[0019] Beneficial effects: The AI-based computing resource intelligent scheduling system and method of the present application uses a resource monitoring module to monitor and predict the use of computing resources, provide a data basis for subsequent decision-making, and ensure that the system can respond to dynamic changes in resources in a timely manner; the multi-dimensional matching module is based on computing power rules and hardware rules, deeply analyzes the multi-dimensional information of the task and matches the most suitable computing center, effectively overcoming the shortcomings of traditional methods in task resource matching; the computing power scheduling module uses simulation scheduling and optimization mechanisms to update the actual scheduling according to the simulation results, greatly improving the scheduling accuracy and resource utilization efficiency, successfully solving the problems of imbalance in multi-task scheduling and insufficient system scalability, significantly enhancing the stability and overall performance of the system, and effectively promoting the improvement of the intelligent level of computing power scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A structural block diagram of an AI-based computing resource intelligent scheduling system provided in an embodiment of the present application; Figure 2 A flowchart of the AI-based intelligent scheduling method for computing resources provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0023] In this article, the term "comprising" is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprising..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0024] The first aspect of this embodiment discloses Figure 1 An AI-based intelligent scheduling system for computing resources is shown, comprising a resource monitoring module, a multi-dimensional matching module and a computing resource scheduling module which are sequentially connected in communication.

[0025] The resource monitoring module is configured to: monitor the real-time usage of computing power resources and generate real-time resource data, predict the usage of computing power resources within a preset prediction period based on the multi-dimensional information of tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information includes at least the computing power requirements and hardware requirements of the task; it should be noted that this embodiment utilizes existing computing center monitoring technology to realize the monitoring of the real-time usage of computing power resources.

[0026] The multi-dimensional matching module is configured as follows: using preset computing power rules and hardware rules, combined with multi-dimensional information, matching the computing center corresponding to the multi-dimensional information, and obtaining the resource allocation amount; wherein the computing power rules correspond to the computing power requirements and enable the matched computing center to meet the computing power requirements, and the hardware rules correspond to the hardware requirements and enable the matched computing center to meet the hardware requirements.

[0027] The computing power scheduling module is configured as follows: performing a first computing power resource scheduling based on the matched computing center, performing a simulated computing power resource scheduling based on the resource prediction data, updating the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and executing after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform a simulated scheduling based on the resource prediction data to obtain the resource optimization amount.

[0028] Based on the above, this embodiment uses the resource monitoring module to monitor and predict the use of computing resources to provide a data basis for subsequent decision-making, ensuring that the system can respond to dynamic changes in resources in a timely manner; the multi-dimensional matching module is based on computing power rules and hardware rules, deeply analyzes the multi-dimensional information of the task and matches the most suitable computing center, effectively overcoming the shortcomings of traditional methods in task resource matching; the computing power scheduling module uses simulation scheduling and optimization mechanisms to update the actual scheduling according to the simulation results, greatly improving the scheduling accuracy and resource utilization efficiency, and successfully solving the problems of imbalance in multi-task scheduling and insufficient system scalability, significantly enhancing the stability and overall performance of the system, and effectively promoting the improvement of the intelligent level of computing power scheduling.

[0029] Specifically, the computing power rules are as follows: Analyze the CPU power requirements of each task , GPU computing power requirements and memory requirements ; Get the available CPU power of each computing center Available GPU computing power , Available Memory And computing power stability index ; Among them, the computing power stability index is used to characterize the stable operation of the computing power of the computing center based on time series; The computing power adaptability of each task assigned to each computing center is calculated based on the CPU computing power requirements, GPU computing power requirements, memory requirements, available CPU computing power, available GPU computing power, available memory, and computing power stability index. The computing center that meets the preset computing power adaptability threshold is defined as the preliminary matching result of the task and output; among which, the computing power adaptability is used to characterize the adaptability of the computing power of the computing center when completing the task.

[0030] As a preferred implementation of this embodiment, this embodiment uses a computing power adaptation degree calculation formula to calculate the computing power adaptation degree, and the computing power adaptation degree calculation formula is specifically: in, The computing power stability index Monitoring cycle in the preset computing center The integral within , and are respectively the weight values ​​corresponding to CPU computing power, GPU computing power and memory set based on common sense known to those skilled in the art, is the minimum value operator, The calculated center To the task The computing power adaptability.

[0031] Based on the above, this embodiment uses a comprehensive analysis of the task computing power requirements and the computing center resources and computing power stability index to achieve accurate computing power adaptation calculation. By analyzing the CPU, GPU and memory requirements of the task in detail, and combining the available resources and computing power stability of the computing center, it is possible to comprehensively and accurately measure the degree of adaptation of each computing center to the task, thereby screening out preliminary matching results that meet the preset threshold. This effectively avoids the rough matching method based only on simple resource quantities and task types in traditional scheduling, greatly improves the matching accuracy between tasks and computing centers at the computing power level, and lays a solid foundation for further optimizing resource allocation in the future, ensuring that tasks can be efficiently executed in the most suitable computing center.

[0032] Specifically, the hardware rules are as follows: Based on different task types Hardware requirements and hardware architecture of the computing center Build a hardware compatibility matrix that stores the compatibility of different task types with hardware architectures It should be noted that this embodiment uses common sense known to those skilled in the art to set the degree of compatibility. This common sense may be, but is not limited to, the correspondence between the hardware architecture and the task type based on the hardware specification.

[0033] Analyze the task type of each task and match the corresponding compatibility level to obtain the hardware stability index of the computing center , the hardware adaptability is calculated based on the compatibility degree and the hardware stability index; among them, the hardware stability index is used to characterize the stable operation of the hardware of the computing center based on the time series, and the hardware adaptability is used to characterize the adaptability of the hardware of the computing center when completing the task.

[0034] The resource allocation amount is calculated based on the hardware adaptability and computing power adaptability, and the preliminary matching results are screened based on the resource allocation amount, and the computing center corresponding to the maximum resource allocation amount is determined as the matched computing center corresponding to the multi-dimensional information and output; wherein the resource allocation amount is used to characterize the computing power and hardware adaptability of the computing center when completing the task.

[0035] It should be noted that the computing power stability index and hardware stability index of this embodiment are both quantitative values ​​based on time series and extracted in combination with the experience known to those skilled in the art that can be used to characterize the characteristics of computing power stability and hardware stability. For example, the characteristics may be, but are not limited to, the failure rate of a computing center.

[0036] As a preferred implementation of this embodiment, this embodiment uses a hardware adaptability calculation formula to calculate the hardware adaptability, and the hardware adaptability calculation formula is specifically: in, Hardware Stability Index Monitoring cycle in the preset computing center The integral within The calculated center (Its hardware architecture is hardware architecture ) for the task (The task type is task type )’s hardware adaptability.

[0037] Further, based on the computing power adaptability calculation formula and the hardware adaptability calculation formula, this embodiment uses the resource allocation amount calculation formula to calculate the resource allocation amount, and the resource allocation amount calculation formula is specifically: in, The weight values ​​of computing power and hardware allocation are set based on common sense known to those skilled in the art. The calculated center To the task The amount of resources allocated.

[0038] Based on the above, this embodiment uses the hardware compatibility matrix and the hardware stability index to achieve accurate calculation of hardware adaptability and determination of the final matching computing center. By constructing a matrix based on the compatibility of different task types with the hardware architecture, and combining it with the hardware stability index, the adaptability of the computing center to the task in terms of hardware can be carefully evaluated. The resource allocation amount is then calculated based on the computing power adaptability, so as to accurately select the optimal computing center in the preliminary matching results. This approach fully takes into account the strict hardware requirements of special tasks, makes up for the shortcomings of traditional methods in hardware matching, and makes the task and the computing center highly compatible at the hardware level, effectively ensuring the efficiency of task execution and avoiding performance degradation caused by hardware mismatch.

[0039] Specifically, based on the result of the simulated computing resource scheduling, the first computing resource scheduling is updated to obtain the second computing resource scheduling, specifically: Get the resource allocation amount of the computing center to which the task is assigned in the first computing power resource scheduling , obtain the resource optimization amount after simulation scheduling ; Based on the resource allocation amount and resource optimization amount, combined with the preset update step coefficient , calculate the update amount, and adjust the resource allocation amount based on the update amount to obtain the second computing power resource scheduling; wherein the update amount is used to optimize the first computing power resource scheduling; it should be noted that the update step coefficient of this embodiment is a parameter obtained by fitting based on the historical update amount.

[0040] As a preferred implementation of this embodiment, this embodiment uses an update amount calculation formula to calculate the update amount, and the update amount calculation formula is specifically: in, is the calculated update amount.

[0041] Based on the above, this embodiment uses the simulated computing power resource scheduling results and the updated step coefficient to achieve effective optimization of the first computing power resource scheduling. By obtaining the resource allocation amount of the original scheduling and the optimization amount after the simulated scheduling, and reasonably calculating the update amount according to the set update step coefficient, the resource allocation is accurately adjusted. This process can make full use of the optimization information provided by the simulated scheduling, make targeted improvements to the initial scheduling plan, effectively overcome the drawbacks of the traditional scheduling that lacks a dynamic adjustment mechanism, make the computing power resource allocation more in line with the actual needs of the task, and significantly improve the system's resource scheduling flexibility and task processing capabilities, ensuring that the system can continuously optimize the scheduling strategy according to actual conditions.

[0042] Specifically, the second computing power resource scheduling is verified and then executed, specifically: Get the first execution time corresponding to the first computing resource scheduling in the computing center , obtain the second execution time corresponding to the second computing resource scheduling in the computing center ; It should be noted that, in this embodiment, the acquisition of the first execution time and the second execution time can be but is not limited to the predicted value based on the historical execution time, and the first execution time can also be a true record of the actual execution time after the first computing power resource is scheduled.

[0043] The task execution efficiency index is calculated based on the resource allocation amount, the resource optimization amount, the first execution time and the second execution time; when the task execution efficiency index is greater than the preset task execution efficiency index threshold, the second computing power resource scheduling verification is determined to be passed, otherwise, the first computing power resource scheduling is updated based on the result of the simulated computing power resource scheduling to obtain the second computing power resource scheduling; wherein, the task execution efficiency index is used to characterize the change in the task execution efficiency of the second computing power resource scheduling compared with the first computing power resource scheduling.

[0044] As a preferred implementation of this embodiment, this embodiment uses a task execution efficiency index calculation formula to calculate the task execution efficiency index, and the task execution efficiency index calculation formula is specifically: in, is the calculated task execution efficiency index. In this embodiment, the task execution efficiency index threshold ,when It is determined that the second computing power resource scheduling verification has passed.

[0045] From the above, this embodiment compares the execution time of the first and second computing power resource scheduling in the computing center, and combines the resource allocation amount to calculate the task execution efficiency index to determine whether the second scheduling scheme is effective. If the threshold is not reached, continue to optimize until the requirements are met. This verification mechanism ensures that each scheduling optimization can effectively improve the efficiency of task execution and avoid invalid or inefficient scheduling adjustments. It effectively solves the problem that traditional scheduling is difficult to evaluate and ensure scheduling effects, provides strong support for the stable and efficient operation of the system, and significantly improves the reliability and practicality of the system.

[0046] Specifically, when scheduling the first computing power resources, if there is a task with a time requirement, the required completion time of the task is obtained. ; Based on the current moment and the estimated execution time of the task with time requirements Determine whether the required completion time is met, if yes (i.e. ) No operation is performed, otherwise (i.e. ) Calculate the resource allocation improvement factor based on the required completion time, the current time and the expected execution time, and recalculate the update amount based on the resource allocation improvement factor; wherein the resource allocation improvement factor is used to optimize the update amount based on the temporal requirements of the task.

[0047] As a preferred implementation of this embodiment, this embodiment uses a resource allocation amount improvement coefficient calculation formula to calculate the resource allocation amount improvement coefficient, and the resource allocation amount improvement coefficient calculation formula is specifically: in, is the calculated resource allocation improvement coefficient. Further, the update amount is optimized to .

[0048] Based on the above, when processing tasks with time requirements, this embodiment makes judgments and calculations based on the current time, the expected execution time, and the required completion time. If the time requirement is not met, the update amount is recalculated by calculating the resource allocation amount improvement coefficient, so as to allocate more suitable resources for such tasks. This effectively avoids the situation where time-sensitive tasks are delayed due to insufficient resources, overcomes the shortcomings of traditional scheduling in processing such tasks, ensures that the system can fully consider the characteristics of various tasks when scheduling, improves the system's adaptability and responsiveness to different task requirements, and ensures that tasks are completed on time.

[0049] Specifically, the simulation computing resource scheduling includes calculating the resource optimization amount obtained after the simulation scheduling; wherein the calculation process of the resource optimization amount is specifically as follows: Get the resource allocation amount of the computing center to which the task is assigned in the first computing power resource scheduling , combined with a preset discount factor and the task transition probability , calculated in the forecast period The simulated resource allocation amount at each time within the period is calculated, the maximum value of the simulated resource allocation amount is obtained, and the maximum value is defined as the resource optimization amount and output.

[0050] It should be noted that the discount factor and the task transfer probability of this embodiment are both empirical values ​​formulated based on common sense known to those skilled in the art.

[0051] As a preferred implementation of this embodiment, this embodiment uses a resource optimization amount calculation formula to calculate the resource optimization amount. The resource optimization amount calculation formula is specifically: in, Indicates that during the forecast period Find the maximum value inside, The calculated center To the task The amount of resource optimization.

[0052] Based on the above, in the simulation of computing resource scheduling, this embodiment combines the resource allocation amount of the first computing resource scheduling and the preset discount factor and task transfer probability to accurately simulate the resource allocation at different times within the prediction period, and determine the resource optimization amount by taking the maximum value. This provides a reliable reference basis for subsequent scheduling optimization, allowing the system to plan resource allocation in advance, effectively respond to dynamic changes in resources and uncertainty in task requirements, improve the system's foresight and the rationality of resource utilization, and improve the scientific nature of scheduling decisions.

[0053] Specifically, when performing the second computing resource scheduling, if there is a large-scale parallel task, the task parallelism of the large-scale parallel task is obtained. ; Obtain the parallel processing capabilities of the computing center , calculate the expected parallel efficiency of the large-scale parallel task in the computing center, update the resource optimization amount based on the expected parallel efficiency and output it; wherein the expected parallel efficiency is used to characterize the processing capability of the computing center for large-scale parallel tasks.

[0054] As a preferred implementation of this embodiment, this embodiment uses an estimated parallel efficiency calculation formula to calculate the estimated parallel efficiency. The estimated parallel efficiency calculation formula is specifically: in, is the minimum value operator, The calculated center Task parallelism The expected parallel efficiency, at this time, the resource optimization amount is optimized to .

[0055] Based on the above, when processing large-scale parallel tasks, this embodiment obtains the task parallelism and the parallel processing capacity of the computing center, calculates the expected parallel efficiency, and updates the resource optimization amount accordingly. This fully considers the special needs of large-scale parallel tasks, makes up for the defects of unreasonable resource allocation in traditional scheduling when processing such tasks, ensures that such tasks can be efficiently executed in the computing center, improves the system's processing ability for complex task types, and further improves the system's resource scheduling performance and task execution effect.

[0056] Specifically, the system further includes an adjustment feedback module that is in communication with the computing power scheduling module, and the adjustment feedback module is configured as follows: After each computing power scheduling is completed, the deviation between the actual data of task execution and the expected data is collected , based on the preset learning rate and the maximum deviation , update the parameter sets in the multi-dimensional matching module and the computing power scheduling module.

[0057] As a preferred implementation of this embodiment, this embodiment uses a parameter update formula to update the parameter set in the multidimensional matching module and the computing power scheduling module. The parameter update formula is specifically: in, For the parameter set in the old multi-dimensional matching module and computing power scheduling module, The parameter sets in the new multi-dimensional matching module and computing power scheduling module are calculated.

[0058] It should be noted that the learning rate in this embodiment is a parameter obtained by adjusting the record fitting based on the historical parameter set.

[0059] Based on the above, this embodiment uses the adjustment feedback module to achieve dynamic updating and optimization of system parameters. After each computing power scheduling is completed, the deviation between the actual task execution data and the expected data is collected, and the parameters in the multidimensional matching module and the computing power scheduling module are intelligently adjusted according to the preset learning rate and maximum deviation value. This enables the system to continuously learn and adapt to actual operating conditions, overcomes the problems of fixed parameters and lack of adaptability of traditional systems, continuously improves the performance and accuracy of the system, ensures that the system is always in an efficient operating state, and enhances the system's adaptability and robustness.

[0060] The second aspect of this embodiment discloses Figure 2An AI-based intelligent scheduling method for computing resources is shown, which is applicable to the AI-based intelligent scheduling system for computing resources as described above, and the method includes: S1: monitor the real-time usage of computing resources and generate real-time resource data, and predict the usage of computing resources within a preset prediction period based on the multi-dimensional information of tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information includes at least the computing power requirements and hardware requirements of the tasks; S2: using preset computing power rules and hardware rules, combined with multi-dimensional information, to match computing centers corresponding to the multi-dimensional information; wherein the computing power rules correspond to computing power requirements and enable the matched computing centers to meet the computing power requirements, and the hardware rules correspond to hardware requirements and enable the matched computing centers to meet the hardware requirements; S3: Perform a first computing power resource scheduling based on the matched computing center, perform a simulated computing power resource scheduling based on the resource prediction data, update the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and execute after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform simulated scheduling based on the resource prediction data.

[0061] It should be noted that the AI-based intelligent scheduling method for computing power resources of this embodiment corresponds to the aforementioned AI-based intelligent scheduling system for computing power resources. Therefore, the contents not specifically described in the AI-based intelligent scheduling method for computing power resources of this embodiment may include but are not limited to function definitions, working principles, and technical effects, etc., and may refer to the records of the aforementioned AI-based intelligent scheduling system for computing power resources, and this text will not elaborate on them here.

[0062] In summary, the AI-based intelligent scheduling system and method for computing resources in this embodiment uses the resource monitoring module to monitor and predict the usage of computing resources, provide a data basis for subsequent decision-making, and ensure that the system can respond to dynamic changes in resources in a timely manner; the multi-dimensional matching module is based on computing power rules and hardware rules, deeply analyzes the multi-dimensional information of the task and matches the most suitable computing center, effectively overcoming the shortcomings of traditional methods in task resource matching; the computing power scheduling module uses simulation scheduling and optimization mechanisms to update the actual scheduling according to the simulation results, greatly improving the scheduling accuracy and resource utilization efficiency, successfully solving the problems of imbalance in multi-task scheduling and insufficient system scalability, significantly enhancing the stability and overall performance of the system, and effectively promoting the improvement of the intelligent level of computing power scheduling.

[0063] In the embodiments provided in the present application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiment can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that can be accessed by a computer. Computer-readable storage media can include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer.

[0064] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An AI-based computing resource intelligent scheduling system, characterized in that: It includes a resource monitoring module, a multi-dimensional matching module and a computing power scheduling module which are sequentially connected in communication; The resource monitoring module is configured to: monitor the real-time usage of computing resources and generate real-time resource data, and predict the usage of computing resources within a preset prediction period based on the multi-dimensional information of the tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information includes at least the computing power requirements and hardware requirements of the tasks; The multi-dimensional matching module is configured to: use the preset computing power rules and hardware rules, in combination with the multi-dimensional information, to match the computing center corresponding to the multi-dimensional information to obtain the resource allocation amount; wherein the computing power rules correspond to the computing power requirements and enable the matched computing center to meet the computing power requirements, and the hardware rules correspond to the hardware requirements and enable the matched computing center to meet the hardware requirements; The computing power scheduling module is configured as follows: performing a first computing power resource scheduling based on the matched computing center, performing a simulated computing power resource scheduling based on the resource prediction data, updating the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and executing after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform a simulated scheduling based on the resource prediction data to obtain a resource optimization amount.

2. The AI-based computing resource intelligent scheduling system according to claim 1 is characterized in that: The computing power rules are specifically as follows: Analyze the CPU computing power requirements, GPU computing power requirements, and memory requirements of each task; Obtain the available CPU computing power, available GPU computing power, available memory and computing power stability index of each computing center; wherein the computing power stability index is used to characterize the stable operation of the computing power of the computing center based on the time series; The computing power adaptation of each task assigned to each computing center is calculated based on the CPU computing power requirement, the GPU computing power requirement, the memory requirement, the available CPU computing power, the available GPU computing power, the available memory and the computing power stability index, and the computing center that meets the preset computing power adaptation threshold is defined as the preliminary matching result of the task and output; wherein the computing power adaptation is used to characterize the degree of adaptation of the computing power of the computing center when completing the task.

3. The AI-based computing resource intelligent scheduling system according to claim 2 is characterized in that: The hardware rules are specifically: A hardware compatibility matrix is ​​constructed based on the hardware requirements of different task types and the hardware architecture of the computing center. The hardware compatibility matrix stores the compatibility degree between different task types and the hardware architecture. Analyze the task type of each task and match the corresponding compatibility level, obtain the hardware stability index of the computing center, and calculate the hardware adaptability based on the compatibility level and the hardware stability index; wherein the hardware stability index is used to characterize the stable operation of the hardware of the computing center based on the time series, and the hardware adaptability is used to characterize the adaptability of the hardware of the computing center when completing the task; The resource allocation amount is calculated based on the hardware adaptability and the computing power adaptability, and the preliminary matching results are screened based on the resource allocation amount, and the computing center corresponding to the maximum value of the resource allocation amount is determined as the matched computing center corresponding to the multi-dimensional information and output; wherein the resource allocation amount is used to characterize the computing power of the computing center and the degree of hardware adaptation when completing the task.

4. The AI-based computing resource intelligent scheduling system according to claim 1 is characterized in that: The updating of the first computing resource scheduling based on the result of the simulated computing resource scheduling to obtain the second computing resource scheduling is specifically: Obtain the resource allocation amount of the computing center to which the task is allocated in the first computing power resource scheduling, and obtain the resource optimization amount obtained after the simulation scheduling; Based on the resource allocation amount and the resource optimization amount, combined with a preset update step coefficient, an update amount is calculated, and the resource allocation amount is adjusted based on the update amount to obtain a second computing power resource scheduling; wherein the update amount is used to optimize the first computing power resource scheduling.

5. The AI-based computing resource intelligent scheduling system according to claim 4 is characterized in that: The verification is performed after the second computing power resource is scheduled, specifically: Obtain a first execution time corresponding to the scheduling of the first computing resource in the computing center, and obtain a second execution time corresponding to the scheduling of the second computing resource in the computing center; The task execution efficiency index is calculated based on the resource allocation amount, the resource optimization amount, the first execution time and the second execution time; when the task execution efficiency index is greater than the preset task execution efficiency index threshold, it is determined that the second computing power resource scheduling verification is passed, otherwise, continue to update the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain the second computing power resource scheduling; wherein, the task execution efficiency index is used to characterize the change in the task execution efficiency of the second computing power resource scheduling compared with the first computing power resource scheduling.

6. The AI-based computing resource intelligent scheduling system according to claim 4 is characterized in that: When performing the first computing power resource scheduling, if there is a task with time requirements, the required completion time of the task is obtained; based on the current time and the estimated execution time of the task with time requirements, it is determined whether the required completion time is met, if so, no operation is performed, otherwise based on the required completion time, the current time and the estimated execution time, the resource allocation amount improvement coefficient is calculated, and the update amount is recalculated based on the resource allocation amount improvement coefficient; wherein, the resource allocation amount improvement coefficient is used to optimize the update amount based on the time requirements of the task.

7. The AI-based computing resource intelligent scheduling system according to claim 1 is characterized in that: The simulated computing power resource scheduling includes calculating the resource optimization amount obtained after the simulated scheduling; wherein the calculation process of the resource optimization amount is specifically as follows: Obtain the resource allocation amount of the computing center to which the task is assigned in the first computing power resource scheduling, combine the preset discount factor and task transfer probability, calculate the simulated resource allocation amount at each time in the prediction period, obtain the maximum value of the simulated resource allocation amount, define the maximum value as the resource optimization amount and output it.

8. The AI-based computing resource intelligent scheduling system according to claim 7 is characterized in that: When performing the second computing power resource scheduling, if there is a large-scale parallel task, the task parallelism of the large-scale parallel task is obtained; the parallel processing capability of the computing center is obtained, the estimated parallel efficiency of the large-scale parallel task in the computing center is calculated, and the resource optimization amount is updated and output based on the estimated parallel efficiency; wherein the estimated parallel efficiency is used to characterize the processing capability of the computing center for large-scale parallel tasks.

9. The AI-based computing resource intelligent scheduling system according to claim 1, characterized in that: The system also includes an adjustment feedback module that is in communication with the computing power scheduling module, and the adjustment feedback module is configured as follows: After each computing power scheduling is completed, the deviation between the actual data of task execution and the expected data is collected, and the parameter sets in the multidimensional matching module and the computing power scheduling module are updated based on the preset learning rate and maximum deviation value.

10. An AI-based intelligent scheduling method for computing resources, the method being applicable to the AI-based intelligent scheduling system for computing resources as claimed in any one of claims 1 to 9, characterized in that: The method includes: S1: monitor the real-time usage of computing resources and generate real-time resource data, predict the usage of computing resources within a preset prediction period based on the multi-dimensional information of tasks in the task queue and generate resource prediction data; wherein the multi-dimensional information at least includes the computing power requirements and hardware requirements of the tasks; S2: using preset computing power rules and hardware rules, combined with the multi-dimensional information, to match a computing center corresponding to the multi-dimensional information; wherein the computing power rules correspond to the computing power requirements and enable the matched computing center to meet the computing power requirements, and the hardware rules correspond to the hardware requirements and enable the matched computing center to meet the hardware requirements; S3: Perform a first computing power resource scheduling based on the matched computing center, perform a simulated computing power resource scheduling based on the resource prediction data, update the first computing power resource scheduling based on the result of the simulated computing power resource scheduling to obtain a second computing power resource scheduling, and execute after verifying the second computing power resource scheduling; wherein, the second computing power resource scheduling is used to optimize the first computing power resource scheduling based on the result of the simulated computing power resource scheduling, and the simulated computing power resource scheduling is used to perform simulated scheduling based on the resource prediction data.

Images