Task priority-based computing power resource allocation method and system

Abstract

This invention relates to the field of computing resource data processing technology, specifically disclosing a computing resource allocation method and system based on task priority. After receiving a target task via the system message bus, the system extracts and verifies multi-dimensional attributes. The verified multi-dimensional attributes of the target task are integrated into a valid dataset, and an initial priority score is generated to classify the target task's priority. The system also determines whether there are any incomplete prerequisite tasks. If there are no prerequisite tasks, the system allocates computing resources and monitors the allocation process parameters. Accurate allocation allows for direct node matching; inaccurate allocation triggers an optimization mechanism to improve resource utilization and execution efficiency. If there are prerequisite tasks, computing resources are pre-allocated, and status monitoring indicators determine whether to adjust the computing resource allocation during execution, thereby reducing potential latency risks. This invention can balance dynamic task changes and dependency constraints, improve the accuracy of computing resource allocation, and optimize overall system performance.

Description

Technical Field

[0001] This invention relates to the field of computing resource data processing technology, and in particular to a computing resource allocation method and system based on task priority. Background Technology

[0002] In existing technologies, task priority-based computing resource allocation methods typically include static priority scheduling, resource demand prediction scheduling, preemptive priority scheduling, and dependency-based scheduling (DAG scheduling). In static priority scheduling, tasks are assigned a fixed priority upon submission, and the scheduler allocates resources sequentially. Resource demand prediction scheduling estimates the required CPU, GPU, and memory resources based on task priorities before allocation. Preemptive scheduling allows high-priority tasks to occupy resources previously held by low-priority tasks. DAG scheduling schedules tasks in topological order using a task dependency graph. Furthermore, some systems incorporate historical execution information or node load status to dynamically fine-tune priorities or resource allocation, while simultaneously monitoring conventional resource metrics such as CPU, GPU, memory, and I / O.

[0003] For example, Chinese invention patent CN117370031A discloses a computing power resource allocation method, device, and system, including a computing power resource allocation method based on a water-filling algorithm. This method includes the following steps: acquiring all available computing power resources and all computing power tasks to be executed; prioritizing the computing power tasks to be executed from high to low according to the strictness of the task completion time requirements; defining a computing power resource allocation optimization objective, which is defined based on maximizing the total computing power task execution rate when all available computing power resources are allocated to each computing power task according to priority; and optimizing the allocation of computing power resources obtained for each computing power task to be executed based on the water-filling algorithm according to the computing power resource allocation optimization objective, thereby completing the allocation of computing power resources.

[0004] For example, Chinese invention patent CN112114967B discloses a GPU resource reservation method based on service priority. The steps are as follows: 1. The GPU obtains the current total GPU resources through relevant configurations, dynamically calculates and reserves the resources required for display according to the current specific display settings, and records the resources currently needed according to the resource application record; 2. The graphics processing task submits graphics commands through a multi-level process: the graphics processing task puts the graphics commands into the first-level command queue, the first-level command queue submits the graphics commands to the second-level command queue according to the priority of the graphics processing task, the second-level command queue submits the graphics commands to the command buffer through the command submission module, and the command buffer submits the graphics commands to the GPU.

[0005] The above-mentioned technologies have at least the following technical problems: Under the existing computing resource allocation technology based on task priority, some tasks may suddenly become performance bottlenecks on the critical path due to the delay of preceding tasks, or the real-time requirements may be higher due to the fluctuation of communication link delay between nodes. The existing solutions are still insufficient in dealing with dynamic changes in tasks and communication constraints, resulting in inaccurate allocation of computing resources, which leads to a decline in overall execution efficiency and limits the improvement of global scheduling performance. Summary of the Invention

[0006] To address the technical problem of insufficient precision in computing resource allocation in existing technologies, this invention provides a method and system for allocating computing resources based on task priority. The technical solution is as follows:

[0007] On the one hand, a method for allocating computing resources based on task priority is provided, which includes:

[0008] After receiving a target task, the system message bus extracts and verifies the multi-dimensional attributes of the target task. The verified multi-dimensional attributes are integrated into a valid dataset, which is then sent to the priority decision engine to generate an initial priority score and assign task priorities. Simultaneously, it determines whether the target task has any incomplete prerequisite tasks. For target tasks without incomplete prerequisite tasks, computing resources are allocated, and the parameters of the allocation process are monitored to determine the accuracy of the allocation results. Based on accurate allocation results, node matching is performed on the target task; for inaccurate allocation results, the computing resource allocation optimization mechanism is triggered. For target tasks with incomplete prerequisite tasks, computing resources are pre-allocated, and the target task's status monitoring indicators are acquired to determine whether the computing resource allocation needs adjustment during execution. This achieves computing resource allocation based on task priority.

[0009] On the other hand, a task priority-based computing resource allocation system is provided. This system includes: a target task extraction module, which, after the system message bus receives a target task, extracts the multi-dimensional attributes of the target task and performs verification processing, integrates the verified multi-dimensional attributes of the target task into a valid dataset, sends the valid dataset to the priority decision engine, thereby generating an initial priority score and dividing the task priority, while determining whether the target task has any incomplete prerequisite dependent tasks; a computing resource allocation module, which allocates computing resources to target tasks without incomplete prerequisite dependent tasks, monitors the parameters of the computing resource allocation process, thereby determining whether the computing resource allocation result of the target task is accurate, performs node matching on the target task based on the accurate computing resource allocation result, and triggers the computing resource allocation optimization mechanism of the target task based on the inaccurate computing resource allocation result; and a computing resource pre-allocation module, which pre-allocates computing resources to target tasks with incomplete prerequisite dependent tasks, obtains the status monitoring indicators of the target task, and determines whether the computing resource allocation needs to be adjusted during the execution of the target task, thereby realizing computing resource allocation based on task priority.

[0010] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0011] (1) This invention provides a method and system for allocating computing resources based on task priority, which can comprehensively consider the multi-dimensional attributes, dependencies, and resource requirements of tasks to achieve accurate allocation and dynamic scheduling of computing resources. By verifying the necessary resources and numerical rationality of tasks, it can ensure that each task has complete critical resources before execution, reducing the risk of task execution failure due to resource shortage; the pre-allocation mechanism based on initial priority score and potential delay risk correction can lock or virtually reserve computing resources in advance, avoid delays in critical path tasks, and improve overall scheduling efficiency; by dynamically adjusting the scheduling window and node allocation using the computing resource allocation accuracy index and status monitoring indicators, it can reduce instantaneous I / O conflicts and computing waste, and optimize the utilization rate of high-load nodes; the node conflict risk sorting and priority-driven preemption mechanism ensure that high-priority tasks obtain available computing power in a timely manner, thereby reducing the risk of task execution delay and improving task execution stability.

[0012] (2) After the system message bus receives the target task, the task parser extracts the multi-dimensional attributes of the task and performs verification processing. The verified attributes are integrated into a valid dataset and sent to the priority decision engine to generate an initial priority score, thereby realizing task priority division. At the same time, it determines whether the task has any incomplete prerequisite dependent tasks. This process ensures that the task submission information is complete and the resource requirements are reasonable, avoiding scheduling failures due to missing key resources or abnormal attributes. At the same time, through the initial priority division, the execution order of high-priority tasks can be reasonably arranged, improving the fairness and efficiency of task scheduling and reducing the overall system waiting time.

[0013] (3) For target tasks without any incomplete prerequisite tasks, the system allocates computing resources and monitors the allocation process parameters to determine the accuracy of the allocation result. For accurate allocation results, node matching is performed directly; for inaccurate allocation results, a computing resource allocation optimization mechanism is triggered. This process can improve resource utilization efficiency, reduce task conflicts between nodes, and enhance the stability of single task execution and global scheduling efficiency. At the same time, the optimization mechanism can dynamically adjust the I / O request volume and scheduling window length of tasks, thereby reducing instantaneous resource pressure and improving the overall system throughput and response speed.

[0014] (4) For target tasks with incomplete prerequisite tasks, the system pre-allocates computing resources, acquires task status monitoring indicators, and determines whether to adjust computing resource allocation based on the deviation between the pending activation status and the execution status indicators. This mechanism enables dynamic resource adjustment, which can lock in the computing power required for critical tasks in advance and reduce execution risks caused by delays in prerequisite tasks or communication fluctuations. In addition, through deviation analysis of status monitoring indicators, the system can predict potential performance bottlenecks and intervene, thereby improving the execution reliability of tasks in the global dependency graph, shortening the overall task completion time, and enhancing the scheduling stability and elasticity of the system in high-load, multi-task environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a computing resource allocation method based on task priority provided in an embodiment of the present invention;

[0017] Figure 2 This is a schematic diagram of the computing resource allocation system based on task priority provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the target task multi-dimensional attribute verification processing flow provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of the computing resource allocation process for the target task provided in an embodiment of the present invention;

[0020] Figure 5 This is a schematic diagram of the target task computing resource allocation optimization mechanism provided in an embodiment of the present invention;

[0021] Figure 6 This is a schematic diagram of the process for pre-allocating computing resources for a target task provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0023] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0024] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0025] This invention provides a method for allocating computing resources based on task priority. For example... Figure 1The flowchart shown is a computing resource allocation method based on task priority. The processing flow of this method may include the following steps: After the system message bus receives the target task, the task parser extracts the multi-dimensional attributes of the target task and performs verification processing. The multi-dimensional attributes of the target task that pass the verification are integrated into a valid dataset. The valid dataset is sent to the priority decision engine to generate an initial priority score and perform task priority division. At the same time, it is determined whether the target task has any unfinished prerequisite dependent tasks. The aforementioned system message bus refers to the communication hub for transmitting information and events within the system. Its main function is to reliably transmit task requests, status information, and control commands, enabling decoupled collaboration and efficient communication among system modules. The aforementioned multi-dimensional attributes of the target task represent a set of characteristics that characterize and quantify the various aspects of the target task, equivalent to multi-angle labels for the target task. Specifically, these include resource requirement attributes, real-time attributes, business value attributes, historical execution attributes, and dependency attributes. The aforementioned determination of whether the target task has any incomplete prerequisite dependent tasks refers to the system traversing the global task dependency graph, searching for the set of prerequisite dependent tasks of the target task, and comparing the execution status of these prerequisite tasks in real time. When any prerequisite task is detected to be incomplete, it is determined that the target task still has incomplete dependencies and is marked as pending activation. If all prerequisite tasks are in a completed state, it is determined that the target task has met the dependency conditions and is marked as ready. The aforementioned prerequisite dependent tasks refer to one or more tasks that the target task must complete before it can begin execution.

[0026] For target tasks without any incomplete prerequisite tasks, computing resources are allocated, and the parameters of the computing resource allocation process are monitored to determine whether the computing resource allocation result of the target task is accurate. Based on the accurate computing resource allocation result, node matching is performed on the target task. Based on the inaccurate computing resource allocation result, the computing resource allocation optimization mechanism of the target task is triggered.

[0027] For target tasks with incomplete prerequisite tasks, computing resources are pre-allocated, and the status monitoring indicators of the target tasks are obtained. It is then determined whether the computing resource allocation needs to be adjusted during the execution of the target tasks, thereby realizing computing resource allocation based on task priority.

[0028] In this invention, to achieve precise allocation of computing resources driven by task priority, a mapping rule system is configured based on historical scheduling data, real-time operation monitoring results, and multi-scenario stress test results. This system integrates the correspondence between multi-dimensional scheduling parameters (such as resource utilization volatility) and multi-level adjustment parameters, and establishes a segmented and scalable parameter mapping table through data modeling and normalization. This mapping rule system not only supports static configuration but can also be dynamically updated based on the computing power allocation accuracy index and state deviation evaluation results, thereby ensuring the real-time adaptability and accuracy of the allocation strategy.

[0029] The system pre-defines various mapping relationships, such as the mapping relationship between time difference and potential latency risk value of the target task, and the mapping relationship between memory allocation fragmentation ratio coefficient, memory bandwidth fluctuation ratio coefficient, and CPU power consumption change rate ratio coefficient and their corresponding effect coefficients. These relationships are all stored in the management database and can be automatically retrieved. The mapping relationships for task priority computing resource allocation involved in this invention can all be obtained through this mapping rule system.

[0030] Specifically, the task parser extracts the multi-dimensional attributes of the target task and performs verification processing. The specific analysis process is as follows: the verification processing includes verification of necessary resources and verification of numerical rationality. Verification of necessary resources refers to obtaining the set of key resources required for the target task to run and determining whether there are any missing file conditions in the set of key resources. The set of key resources refers to the set of resource items necessary for the target task to execute, specifically including configuration files, dependency library files, runtime scripts, data input files, hardware driver files, etc. If there are missing file conditions in the set of key resources, a task rejection instruction is generated and returned to the task initiator for supplementary explanation. If there are no missing file conditions in the set of key resources, numerical rationality verification is performed. The aforementioned missing file condition refers to any resource item in the set of key resources being missing in the target task description file.

[0031] Numerical rationality verification is performed, specifically as follows: The resource requirement parameter set of the target task is obtained and compared with the numerical constraint rule table to determine if there are any numerical anomalies in the target task's resource requirement parameter set. The resource requirement parameter set of the target task describes the set of numerical values ​​corresponding to the computing resources required by the target task during execution, specifically including CPU resource parameters, GPU resource parameters, network resource parameters, etc. The numerical constraint rule table is used to limit the values ​​corresponding to any resource in the target task's resource requirement parameter set to within the executable range, specifically including CPU resource constraint rules, GPU resource constraint rules, network resource constraint rules, etc. Numerical anomalies... A normal condition refers to a value corresponding to any resource in the resource requirement parameter set of the target task exceeding the constraint range corresponding to the numerical constraint value rule table. If there are abnormal numerical conditions in the resource requirement parameter set of the target task, the abnormal values ​​corresponding to the abnormal resources in the resource requirement parameter set of the target task are directly adjusted to the boundary values ​​in the numerical constraint value rule table, and integrated into a valid dataset. If there are no abnormal numerical conditions in the resource requirement parameter set of the target task, the resource requirement parameter set of the target task remains unchanged, and is integrated into a valid dataset. The boundary values ​​in the aforementioned numerical constraint value rule table include upper and lower limits, which are used to limit the resource requirement parameters of the target task within the executable range. When there are abnormal numerical conditions in the resource requirement parameter set of the target task, the system will select the applicable boundary value according to the specific abnormal situation: if the value corresponding to any resource in the resource requirement parameter set of the target task exceeds the upper limit, it is adjusted to the corresponding upper limit value; if the value corresponding to any resource in the resource requirement parameter set of the target task is lower than the lower limit, it is adjusted to the corresponding lower limit value, thereby ensuring that the resource parameters are within the system's acceptable and executable range, and are integrated into a valid dataset for subsequent task scheduling and computing resource allocation.

[0032] In one specific embodiment, when the resource requirement parameter set of the target task includes a memory request value, if the memory requirement submitted by the target task is 256GB, while the maximum allowed memory value preset in the system's numerical constraint rule table is 128GB, then this memory requirement is considered an abnormal value. In this case, the system will automatically adjust the abnormal value to the upper limit boundary value of 128GB in the rule table, thereby avoiding task scheduling failures or resource waste caused by excessive memory requests.

[0033] In one specific embodiment, if the scheduling time window of the target task is set to 0.5 seconds, and the minimum schedulable time window specified in the numerical constraint value rule table is 1 second, the system will automatically adjust it to the boundary lower limit value of 1 second after detecting that the time window is lower than the lower limit, so as to ensure that the task has a minimum executable time and avoid the task failing to start or execute normally due to the scheduling window being too short.

[0034] like Figure 3 As shown in the schematic diagram of the target task multi-dimensional attribute verification process provided in this embodiment of the invention, after the system message bus receives the target task, the task parser verifies the multi-dimensional attributes of the target task. First, it checks the necessary resources of the task to determine whether there are any missing file conditions for key resources; if there are missing conditions, a task rejection instruction is generated; if there are no missing conditions, it checks the numerical rationality to determine whether there are any abnormal numerical conditions in the resource requirement parameter set. If there are abnormal numerical conditions, the abnormal values ​​are adjusted to the boundary values ​​in the numerical constraint rule table; if there are no abnormal conditions, they are integrated into a valid dataset and an initial priority score is generated. Then it is determined whether the initial priority score is lower than the priority score threshold to distinguish high-priority tasks from non-high-priority tasks.

[0035] Specifically, the process of generating initial priority scores and assigning task priorities involves the following steps: Generating initial priority scores refers to accumulating scores based on a priority score mapping table. The priority score mapping table defines the correspondence between the values ​​of each computing resource in the valid dataset and their priority scores. The aforementioned score accumulation process based on the priority score mapping table involves the system reading the values ​​of each computing resource in the valid dataset of the target task, such as the number of CPU cores, the number of GPU cores, video memory capacity, memory size, I / O bandwidth, and runtime window. For each computing resource, the system searches for the corresponding priority score in the priority score mapping table and accumulates the corresponding priority scores of each computing resource in the valid dataset to obtain the total score, which is the initial priority score of the target task.

[0036] Task priority classification refers to comparing the initial priority score with the priority score threshold to determine whether the target task is a high-priority task. The aforementioned priority score threshold refers to the minimum initial priority score required for high-priority tasks preset in the management database. If the initial priority score is not lower than the priority score threshold, the target task is determined to be a high-priority task; if the initial priority score is lower than the priority score threshold, the target task is determined not to be a high-priority task.

[0037] Furthermore, for target tasks without any incomplete prerequisite tasks, computing resources are allocated. The specific analysis process is as follows:

[0038] Computing resource allocation refers to allocating computing resources based on initial priority scores. Specifically, high-priority tasks are allocated from a fixed computing resource reservation pool, while non-high-priority tasks are allocated from a dynamic computing resource pool. The fixed computing resource reservation pool is exclusively for high-priority tasks, and its resources cannot be preempted, ensuring the timely execution of high-priority tasks. The dynamic computing resource pool is for non-high-priority tasks, supporting preemption and sharing; high-priority tasks can temporarily occupy the computing resources of non-high-priority tasks. Specifically, the system allocates the actual amount of computing resources to the target task based on its resource requirement parameter set and priority. For high-priority tasks, the allocated computing resources equal all critical computing resources required during execution, ensuring timely and stable task completion. For non-high-priority tasks, the allocated computing resources are determined based on the available resources in the dynamic computing resource pool. The system allocates resources based on available demand, either all or a portion of it, supporting preemption or sharing during resource shortages to balance resource utilization efficiency and task execution fairness. The resource requirement parameter set for the target task is a portion of the effective dataset, representing core information about computing resource allocation, including specific requirements for CPU / GPU / memory / bandwidth / runtime windows. The system also monitors parameters related to the computing resource allocation process, including memory allocation fragmentation ratios, I / O channel availability ratios, and historical task execution interference ratios. By pre-setting effect coefficients for each ratio in the management database, the system quantifies their weighted contribution to the computing resource allocation accuracy index. Finally, a weighted average fusion algorithm is used to synthesize the computing resource allocation accuracy index, which represents the degree to which the actual allocated resources match the task requirements when allocating computing resources to the target task.

[0039] The memory allocation fragmentation ratio coefficient represents the ratio of memory allocation fragmentation to the defined memory allocation fragmentation. The I / O channel availability index ratio coefficient represents the ratio of I / O channel availability index to the defined I / O channel availability index. The historical task execution interference index ratio coefficient represents the ratio of historical task execution interference index to the defined historical task execution interference index.

[0040] The aforementioned memory allocation fragmentation is used to measure the obstacles that the target task may encounter when requesting memory. It is obtained by reading the node memory allocation status at the resource management system level and calculating the ratio of total available memory to the maximum contiguous available block size. The aforementioned I / O channel availability index reflects the idle bandwidth and availability of the network I / O channel. It is obtained by monitoring the node's bandwidth utilization, I / O request queue length, and average latency, and then normalizing and weighting each index. The specific weighting weights are determined by those skilled in the art based on the actual situation. The aforementioned historical task execution interference index represents the potential competition between the target task and running tasks in terms of computing resources. It is obtained by normalizing and weighting each index based on the resource utilization (such as CPU utilization), concurrency overlap (such as overlap time ratio), latency fluctuation (such as queuing time fluctuation), and blocking event statistics (such as page swapping due to insufficient memory) of historical tasks. The specific weighting weights are determined by those skilled in the art based on the actual situation.

[0041] The above definition of memory allocation fragmentation represents the maximum value of memory allocation fragmentation within the specified range; the above definition of I / O channel availability index represents the minimum value of I / O channel availability index within the specified range; the above definition of historical task execution interference index represents the maximum value of historical task execution interference index within the specified range.

[0042] In the process of computing resource allocation, the larger the ratio of memory allocation fragmentation to the defined memory allocation fragmentation, the larger the memory allocation fragmentation ratio coefficient, indicating that the node's memory fragmentation is more severe. This will reduce memory utilization efficiency and increase task scheduling interference, thereby affecting the effective allocation of other resources and causing a decrease in the accuracy index of computing resource allocation. The larger the I / O channel availability ratio coefficient, the smoother the I / O channel, which can support more concurrent tasks, improve task execution stability, and thus mitigate the impact of memory and historical interference on resource allocation, thereby improving the accuracy index of computing resource allocation. The larger the historical task execution interference ratio coefficient, the more severe the historical task interference on the node, which may cause task performance fluctuations and resource competition, thus causing a decrease in the accuracy index of computing resource allocation.

[0043] When memory fragmentation is high, the number of contiguous available memory blocks decreases, leading to frequent I / O operations by tasks and thus affecting I / O channel availability metrics. Simultaneously, interference from historical task executions may amplify this impact, as high-concurrency or overlapping tasks can consume memory and I / O resources, making the negative effects of memory fragmentation and I / O availability more pronounced. Conversely, when I / O channel availability is high, the time tasks spend waiting for I / O is reduced, alleviating the pressure on resource allocation from memory fragmentation and historical task interference, thereby improving the overall accuracy of computing resource allocation.

[0044] The specific evaluation method for the accuracy index of computing resource allocation is as follows:

[0045]

[0046] In the formula, CRA is the accuracy index of computing resource allocation, MDFC is the memory allocation fragmentation ratio coefficient in the computing resource allocation process, IOAC is the I / O channel availability index ratio coefficient in the computing resource allocation process, HTIC is the historical task execution interference index ratio coefficient in the computing resource allocation process, fg is the effect coefficient corresponding to the memory allocation fragmentation ratio coefficient preset in the management database, fm is the effect coefficient corresponding to the I / O channel availability index ratio coefficient preset in the management database, and fc is the effect coefficient corresponding to the historical task execution interference index ratio coefficient preset in the management database.

[0047] The effect coefficients corresponding to the memory allocation fragmentation ratio coefficients mentioned above indicate that when the memory allocation fragmentation ratio coefficient changes by a unit magnitude, the computing resource allocation accuracy index will change accordingly. The effect coefficients corresponding to the I / O channel availability index ratio coefficients mentioned above indicate that when the I / O channel availability index ratio coefficient changes by a unit magnitude, the computing resource allocation accuracy index will change accordingly. The effect coefficients corresponding to the historical task execution interference index ratio coefficients mentioned above indicate that when the historical task execution interference index ratio coefficient changes by a unit magnitude, the computing resource allocation accuracy index will change accordingly.

[0048] The management database stores the mapping relationships between memory allocation fragmentation ratio coefficients and their corresponding effect coefficients, I / O channel availability index ratio coefficients and their corresponding effect coefficients, and historical task execution interference index ratio coefficients and their corresponding effect coefficients. In this embodiment, the mapping relationship is a mapping table. For example, by inputting the memory allocation fragmentation ratio coefficient, I / O channel availability index ratio coefficient, and historical task execution interference index ratio coefficient into the management database, the management data can match the effect coefficients corresponding to the memory allocation fragmentation ratio coefficient, the I / O channel availability index ratio coefficient, and the historical task execution interference index ratio coefficient based on the preset mapping relationship. The numerical range of each effect coefficient is strictly controlled between 0 and 1.

[0049] Specifically, the process for determining the accuracy of the computing resource allocation result for the target task is as follows: The computing resource allocation accuracy index is compared with the computing resource allocation accuracy threshold, which represents the minimum value of the computing resource allocation accuracy index within a specified range preset in the management database. If the computing resource allocation accuracy index is greater than or equal to the computing resource allocation accuracy threshold, the computing resource allocation result for the target task is deemed accurate, and node matching for the target task is performed. If the computing resource allocation accuracy index is less than the computing resource allocation accuracy threshold, the computing resource allocation result for the target task is deemed inaccurate, and the computing resource allocation optimization mechanism for the target task is triggered.

[0050] Furthermore, node matching is performed for the target task. The specific analysis process is as follows: it is determined whether there are nodes in the current node cluster that meet the computing power resource requirements of the target task. Specifically, the system obtains the available resource information of each node from the cluster management database or monitoring module, including the number of CPU cores, GPU cores and memory capacity, memory capacity, network bandwidth, I / O channel availability, etc. The resource requirement parameter set of the target task is compared with the available resources of each node item by item to determine whether the available resources of the nodes meet the requirements of the target task in terms of CPU, GPU, memory, bandwidth and runtime window.

[0051] If there are nodes in the current node cluster that meet the computing power resource requirements of the target task, these nodes are marked as valid nodes. The conflict risk value of each valid node is obtained and sorted in ascending order of conflict risk value. The valid node with the highest conflict risk value is selected for binding to the target task. Specifically, when matching nodes for the target task, the Top-Approximation Ideal Solution Ranking Method (TOPSIS) is used to assess node conflict risk. This involves: first, constructing a multi-dimensional conflict index matrix for each valid node, including resource utilization, historical task interference index, and communication constraints, and normalizing each index; then generating a weighted matrix based on preset weights to reflect the relative importance of each index to node conflict risk; subsequently, the system further determines... Ideal and negative ideal solutions are defined, where the ideal solution represents the vector of optimal values ​​(maximum or minimum, depending on the nature of the indicator) among the weighted indicators for each column, and the negative ideal solution represents the vector of worst values ​​(maximum or minimum, depending on the nature of the indicator) among the weighted indicators for each column. The Euclidean distances from each valid node to the ideal solution and to the negative ideal solution are calculated. The Top-Ideal-Solution Ranking (TOPSIS) method is used to convert the Euclidean distances into a relative proximity index. This relative proximity index reflects the degree to which each valid node approaches the ideal state; a larger index indicates a lower conflict risk for each valid node. Finally, the relative proximity indices of each valid node are linearly transformed to obtain the conflict risk value for each valid node. This linear transformation involves subtracting the relative proximity index of each valid node from 1; the result is the conflict risk value for each valid node.

[0052] If there are no nodes in the current node cluster that meet the computing power resource requirements of the target task, the task will be classified and processed based on its priority. If the target task is of high priority, the computing power resource preemption mechanism will be triggered, and the computing power resources of non-high priority tasks will be temporarily occupied. If the target task is not of high priority, the task will wait for idle computing power resources.

[0053] like Figure 4As shown in the schematic diagram of the target task computing resource allocation process provided in this embodiment of the invention, the system determines whether the target task has any incomplete prerequisite tasks. If there are incomplete prerequisite tasks, computing resources are pre-allocated to the target task; if there are no incomplete prerequisite tasks, computing resources are allocated, and then the computing resource allocation accuracy index is obtained and it is determined whether it is not lower than the computing resource allocation accuracy threshold. If it is lower than the computing resource allocation accuracy threshold, the computing resource allocation optimization mechanism for the target task is triggered; otherwise, node matching is performed on the target task: it is determined whether there are any nodes in the current node cluster that meet the computing resource requirements of the target task. If there are any nodes that meet the computing resource requirements of the target task, the nodes that meet the requirements are marked as valid nodes, and the conflict risk value of each valid node is obtained. The valid node with the highest conflict risk value is selected for binding to the target task; if there are no nodes that meet the computing resource requirements of the target task, the target task is classified and processed based on its priority. High-priority tasks trigger the computing resource preemption mechanism, while non-high-priority tasks wait for idle computing resources.

[0054] Specifically, the optimization mechanism for the allocation of computing resources for the target task is triggered by the following analysis process: Based on the accuracy index and the accuracy threshold of the allocation of computing resources, the accuracy deviation value is obtained; based on the accuracy deviation value, the reduction amount of I / O requests for a single task is matched to reduce the instantaneous occupation of the I / O channel by a single task during execution; based on the accuracy deviation value, the scheduling window duration is increased by a coefficient, thereby increasing the scheduling window duration of the target task.

[0055] The aforementioned acquisition of the allocation accuracy deviation value refers to subtracting the computing resource allocation accuracy index from the computing resource allocation accuracy limit value. The specific matching process for determining the reduction in single-task I / O requests based on the allocation accuracy deviation value is as follows: the management database stores a mapping table of allocation accuracy deviation value and single-task I / O request reduction. The allocation accuracy deviation value is input into the management database, and the management database, based on this mapping table, can match the single-task I / O request reduction amount. Subtracting the matched single-task I / O request reduction amount from the original single-task I / O request amount yields the required adjustment to the single-task I / O request amount. Reducing the single-task I / O request amount based on the allocation accuracy deviation value can reduce the instantaneous bandwidth consumption and I / O queue waiting time during single-task execution, thereby alleviating resource contention and improving the availability and matching degree of computing resources.

[0056] The above-mentioned matching of the scheduling window duration increase factor based on the allocation accuracy deviation value is as follows: The management database presets the scheduling window duration increase factor corresponding to each allocation accuracy deviation value interval. The obtained allocation accuracy deviation value is input into the management database, and the management database can match the corresponding allocation accuracy deviation value interval. The scheduling window duration increase factor corresponding to this interval is the required increase factor. The obtained scheduling window duration increase factor is multiplied by the original scheduling window duration, and the result is the scheduling window duration that needs to be adjusted. The above-mentioned scheduling window duration increase factor is greater than 1, which indicates that the scheduling window duration of the target task needs to be increased by a certain factor. By increasing the scheduling window duration, the scheduler can take into account the needs of more tasks when allocating computing resources, thereby achieving a more reasonable allocation of computing resources and load balancing. Ultimately, this helps to improve the accuracy index of computing resource allocation and improve the execution efficiency and stability of the overall scheduling system.

[0057] After optimizing the allocation of computing resources, the accuracy index of computing resource allocation is re-acquired and marked as the final value of the accuracy of computing resource allocation, so as to determine whether to issue an information warning for the computing resource allocation process of the target task.

[0058] Furthermore, the system determines whether to issue an information warning for the computing resource allocation process of the target task. The specific determination process is as follows: the accurate final value of computing resource allocation is compared with the accurate limit value of computing resource allocation; if the accurate final value of computing resource allocation is greater than or equal to the accurate limit value of computing resource allocation, it is determined that no information warning will be issued for the computing resource allocation process of the target task; if the accurate final value of computing resource allocation is less than the accurate limit value of computing resource allocation, it is determined that an information warning will be issued for the computing resource allocation process of the target task. Specifically, an warning instruction is generated and a warning log is recorded. The warning information is sent to the scheduling management module and the operation and maintenance monitoring module to prompt the operation and maintenance personnel to pay attention to the resource allocation status of the current task.

[0059] like Figure 5 As shown in the flowchart of the target task computing resource allocation optimization mechanism provided in this embodiment of the invention, when the target task computing resource allocation accuracy index does not reach the specified limit, the computing resource allocation optimization mechanism is triggered. First, the allocation accuracy deviation value is obtained, and the single-task I / O request volume is adjusted according to the deviation value to reduce the instantaneous occupation of the I / O channel by the target task. Simultaneously, a matching scheduling window increase coefficient is applied to extend the target task scheduling window duration. After adjustment, the final value of the computing resource allocation accuracy is obtained, and it is determined whether the final value is not lower than the computing resource allocation accuracy limit value. If it is not lower than the computing resource allocation accuracy limit value, no warning is issued; if it is lower than the computing resource allocation accuracy limit value, a warning is triggered.

[0060] Specifically, computing resources are pre-allocated for target tasks with incomplete prerequisite dependent tasks. The specific analysis process is as follows:

[0061] Pre-allocation of computing resources refers to the pre-allocation of computing resources based on a modified initial priority score. Specifically, high-priority tasks lock computing resources in advance from a fixed computing resource reservation pool, while non-high-priority tasks virtually lock resources from a dynamic computing resource pool, marking them as reserved resources for the future and not immediately occupied. The modified initial priority score refers to obtaining the earliest and latest start times of the target task, obtaining the time difference based on the deviation between the latest and earliest start times, matching the potential delay risk value of the target task based on the time difference, and modifying the initial priority score based on the potential delay risk value of the target task, thereby obtaining... The corrected initial priority score; the earliest start time of the aforementioned target task refers to the earliest time when all prerequisite tasks of the target task are completed. It is obtained by: the system iterating through each prerequisite task of the target task, obtaining the completion time of each prerequisite task, and selecting the latest completion time as the earliest start time of the target task, ensuring that the target task can only start after all prerequisite tasks are completed; the latest start time of the aforementioned target task refers to the latest time the target task can start without affecting the final completion time of the entire global task dependency graph. It is obtained by: the system first determining the critical path where the target task is located based on the global task dependency graph, and obtaining the critical path. The final completion time is calculated; then, following the order of tasks on the critical path, the estimated execution time of each task is subtracted sequentially from the final completion time of the critical path, until the estimated execution time of the target task itself is subtracted, to calculate the latest start time of the target task. This ensures that if the target task starts execution at this time, it will not delay the final completion time of the entire critical path. The above-mentioned potential delay risk value of the target task is matched based on the time difference. The specific matching process is as follows: the management database stores a mapping relationship table between the time difference and the potential delay risk value of the target task. The time difference is input into the management database, and the management database can match the corresponding potential delay risk value of the target task. The initial priority score is corrected based on the potential delay risk value of the target task, specifically as follows: P = P0 × (1 + α × R), where P is the corrected initial priority score, P0 is the initial priority score, R is the potential delay risk value of the target task, and α is an adjustment coefficient set by professionals in the field to control the degree of influence of risk in the correction. Substituting the obtained potential delay risk value of the target task into this formula yields the corrected initial priority score. The specific correction principle is: the larger the potential delay risk value of the target task, the larger the corrected initial priority score, so that the target task obtains higher priority in scheduling, thereby allocating computing resources in advance to reduce the possibility of delay.

[0062] Simultaneously, the system acquires status monitoring indicators for the pre-allocation of computing resources for the target task. Specifically, during the pre-allocation phase, the system collects various computing resource allocation data in real time from the management database and scheduling system, taking the target task into an inactive state. This data is then normalized according to a preset ratio coefficient to generate status monitoring indicators, which are marked as inactive state indicators. Inactive state indicators refer to predicted indicators when the target task is not executed. Similarly, the system acquires status monitoring indicators during target task execution. During the execution phase, the system also collects computing resource allocation data and generates status monitoring indicators using the same method, marking them as active state indicators. Active state indicators refer to actual indicators during target task execution. Both types of status indicators are automatically updated through the real-time monitoring module to ensure data accuracy and availability. These status monitoring indicators are used to assess the computing power status of the target task. The aforementioned preset ratio coefficient is a pre-set ratio coefficient in the management database. The aforementioned computing resource allocation data includes, but is not limited to, CPU resources, GPU resources, and memory resources.

[0063] Based on the pending and activated status indicators, the deviation value of the target task status indicator is obtained to determine whether the allocation of computing resources needs to be adjusted during the execution of the target task. The specific judgment process is as follows: the deviation value of the target task status indicator is compared with the deviation limit value of the target task status indicator. The deviation limit value of the target task status indicator represents the maximum value of the target task status indicator deviation value preset in the management database within a specified range. The above-mentioned acquisition of the target task status indicator deviation value refers to the absolute value of subtracting the pending status indicator from the activated status indicator. By measuring the difference between the activated status indicator and the pending status indicator, the degree of deviation between the actual execution of the target task and the expected plan is intuitively reflected, thereby providing a basis for subsequent judgment on whether the allocation of computing resources needs to be adjusted.

[0064] If the deviation value of the target task status indicator is greater than or equal to the target task status indicator deviation limit, it is determined that the target task needs to adjust the allocation of computing resources during execution. Specifically, this includes, but is not limited to, dynamically recalculating the target task priority and optimizing node selection. Dynamically recalculating the target task priority refers to re-acquiring the initial priority score and dynamically updating its sorting position in the global task scheduling queue based on the initial priority score. Optimizing node selection refers to obtaining all valid nodes in the current node cluster that meet the computing resource requirements of the target task and selecting the valid node with the lowest conflict risk value.

[0065] If the deviation value of the target task status index is less than the target task status index deviation limit value, it is determined that the target task does not need to adjust the allocation of computing resources during execution.

[0066] like Figure 6As shown in the schematic diagram of the target task computing resource pre-allocation process provided in this embodiment of the invention, the potential delay risk value of the target task is matched based on the time difference, and the initial priority score is corrected accordingly. Then, computing resources are pre-allocated based on the corrected initial priority score. After that, based on the pending activation status index and the activation status index of the target task computing resources pre-allocation, the target task status index deviation value is obtained. It is determined whether the target task status index deviation value is not lower than the target task status index deviation limit value. If it is not lower than the target task status index deviation limit value, no adjustment of computing resource allocation is required during execution; otherwise, the computing resource allocation needs to be adjusted during execution.

[0067] A second aspect of this invention provides a computing resource allocation system based on task priority, such as... Figure 2 The diagram shows the structure of a task priority-based computing resource allocation system, which includes: a target task extraction module, a computing resource allocation module, a computing resource pre-allocation module, and a management database.

[0068] The target task extraction module is connected to the computing resource allocation module and the computing resource pre-allocation module, respectively. The target task extraction module, the computing resource allocation module, and the computing resource pre-allocation module are all connected to the management database. The management database is used to store various parameters involved in the computing resource allocation system based on task priority.

[0069] The target task extraction module is used to extract multi-dimensional attributes of the target task after the system message bus receives the target task, and to perform verification processing. The multi-dimensional attributes of the target task that pass the verification are integrated into a valid dataset, which is then sent to the priority decision engine to generate an initial priority score and divide the task priority. At the same time, it is determined whether the target task has any unfinished prerequisite tasks. The computing resource allocation module is used to allocate computing resources to target tasks without unfinished prerequisite tasks, monitor the parameters of the computing resource allocation process, and determine whether the computing resource allocation result of the target task is accurate. Based on the accurate computing resource allocation result, node matching is performed on the target task. Based on the inaccurate computing resource allocation result, the computing resource allocation optimization mechanism of the target task is triggered. The computing resource pre-allocation module is used to pre-allocate computing resources to target tasks with unfinished prerequisite tasks, and to obtain the status monitoring indicators of the target task. It also determines whether the computing resource allocation needs to be adjusted during the execution of the target task, thereby realizing computing resource allocation based on task priority.

[0070] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0071] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0072] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for allocating computing resources based on task priority, characterized in that, The method includes: After the system message bus receives the target task, the task parser extracts the multi-dimensional attributes of the target task and performs verification processing. The verified multi-dimensional attributes of the target task are integrated into a valid dataset, and the valid dataset is sent to the priority decision engine to generate an initial priority score and divide the task priority. At the same time, it is determined whether the target task has any unfinished prerequisite dependent tasks. For target tasks without any incomplete prerequisite tasks, computing resources are allocated, and the parameters of the computing resource allocation process are monitored to determine whether the computing resource allocation result of the target task is accurate. Based on the accurate computing resource allocation result, node matching is performed on the target task. Based on the inaccurate computing resource allocation result, the computing resource allocation optimization mechanism of the target task is triggered. The specific analysis process for allocating computing resources to target tasks that do not have any incomplete prerequisite tasks is as follows: The aforementioned allocation of computing resources refers to allocating computing resources based on initial priority scores and monitoring parameters during the allocation process. The parameters of the computing resource allocation process include memory allocation fragmentation ratio coefficient, I / O channel availability index ratio coefficient, and historical task execution interference index ratio coefficient. By presetting the effect coefficient of each ratio coefficient in the management database, the weight contribution value of each ratio coefficient to the computing resource allocation accuracy index is quantified. Finally, a weighted average fusion algorithm is used to synthesize the computing resource allocation accuracy index, which represents the degree to which the actual allocated resources match the task requirements when the system allocates computing resources to the target task. The memory allocation fragmentation ratio coefficient represents the ratio of memory allocation fragmentation to the defined memory allocation fragmentation; the I / O channel availability index ratio coefficient represents the ratio of I / O channel availability index to the defined I / O channel availability index; and the historical task execution interference index ratio coefficient represents the ratio of historical task execution interference index to the defined historical task execution interference index. The specific analysis process of the computing resource allocation optimization mechanism that triggers the target task is as follows: Based on the accuracy index and the accuracy limit of computing resource allocation, the allocation accuracy deviation value is obtained. Based on the allocation accuracy deviation value, the reduction amount of I / O requests for a single task is matched, thereby reducing the instantaneous occupation of the I / O channel by a single task during execution. Based on the allocation accuracy deviation value, the scheduling window duration increase coefficient is matched, thereby increasing the scheduling window duration of the target task. After optimizing the allocation of computing resources, the accuracy index of computing resource allocation is re-acquired and marked as the final value of the accuracy of computing resource allocation, so as to determine whether to issue an information warning for the computing resource allocation process of the target task. For target tasks with incomplete prerequisite tasks, computing resources are pre-allocated, and the status monitoring indicators of the target tasks are obtained. It is also determined whether the computing resource allocation needs to be adjusted when the target tasks are executed, thereby realizing computing resource allocation based on task priority. The specific analysis process for pre-allocating computing resources for target tasks with incomplete prerequisite dependent tasks is as follows: The aforementioned pre-allocation of computing resources refers to the pre-allocation of computing resources based on the corrected initial priority score. The corrected initial priority score refers to obtaining the earliest start time and the latest start time of the target task, obtaining the time difference value based on the deviation between the latest start time and the earliest start time of the target task, matching the potential delay risk value of the target task based on the time difference value, and correcting the initial priority score based on the potential delay risk value of the target task to obtain the corrected initial priority score. Simultaneously, the status monitoring indicators for the pre-allocation of computing resources for the target task are acquired and marked as indicators to be activated. The status monitoring indicators for the execution of the target task are also acquired and marked as indicators to be activated. These status monitoring indicators are used to evaluate the computing power status of the target task. Based on the pending and activated status indicators, the deviation value of the target task status indicator is obtained to determine whether the computing power resource allocation needs to be adjusted during the execution of the target task. The specific judgment process is as follows: the deviation value of the target task status indicator is compared with the deviation limit value of the target task status indicator. The deviation limit value of the target task status indicator represents the maximum value of the deviation value of the target task status indicator within the specified range. If the deviation value of the target task status index is greater than or equal to the deviation limit value of the target task status index, it is determined that the target task needs to adjust the allocation of computing resources during execution. If the deviation value of the target task status index is less than the target task status index deviation limit value, it is determined that the target task does not need to adjust the allocation of computing resources during execution.

2. The method of claim 1, wherein, The task parser extracts the multi-dimensional attributes of the target task and performs verification processing. The specific analysis process is as follows: The verification process includes verification of necessary resources for the task and verification of the reasonableness of the values. The necessary resource verification for the task refers to obtaining the set of key resources required for the target task to run and determining whether there are any missing files in the set of key resources. The set of key resources refers to the set of resource items required by the target task during execution. If the critical resource set has missing files, a task rejection instruction will be generated and returned to the task initiator for supplementary explanation. If the set of critical resources does not have any missing files, then a numerical validity check will be performed. The missing file condition refers to the absence of any resource item in the key resource set in the target task description file. The numerical reasonableness verification process is as follows: Obtain the resource requirement parameter set of the target task and compare it with the numerical constraint value rule table to determine whether there are any numerical anomalies in the resource requirement parameter set of the target task. The resource requirement parameter set of the target task is used to describe the set of numerical values ​​corresponding to the computing power resources required by the target task during execution. The numerical constraint value rule table is used to limit the numerical values ​​corresponding to any resource in the resource requirement parameter set of the target task to be within the executable range. The numerical abnormal condition refers to the numerical value corresponding to any resource in the resource requirement parameter set of the target task exceeding the constraint range corresponding to the numerical constraint value rule table. If there are numerical anomalies in the resource requirement parameter set of the target task, the abnormal values ​​corresponding to the abnormal resources in the resource requirement parameter set of the target task will be directly adjusted to the boundary values ​​in the numerical constraint value rule table, and then integrated into a valid dataset. If the resource requirement parameter set of the target task does not have any numerical anomalies, then the resource requirement parameter set of the target task remains unchanged and is integrated into a valid dataset.

3. The method of claim 1, wherein, The specific analysis process for generating initial priority scores and assigning task priorities is as follows: Generating an initial priority score refers to accumulating scores based on a priority score mapping table to obtain an initial priority score. The priority score mapping table is used to define the correspondence between the numerical values ​​of each computing resource in the effective dataset and the priority scores. The process of prioritizing tasks refers to comparing the initial priority score with the priority score threshold to determine whether the target task is a high-priority task. If the initial priority score is not lower than the priority score threshold, the target task is determined to be a high-priority task. If the initial priority score is lower than the priority score threshold, the target task is determined not to be a high-priority task.

4. The method of claim 1, wherein, The specific process for determining the accuracy of the target task's computing resource allocation result is as follows: The accuracy index of computing resource allocation is compared with the accuracy limit value of computing resource allocation, wherein the accuracy limit value of computing resource allocation represents the minimum value of the accuracy index of computing resource allocation within a specified range. If the accuracy index of computing power resource allocation is greater than or equal to the accuracy threshold of computing power resource allocation, then the computing power resource allocation result of the target task is judged to be accurate, and node matching is performed on the target task. If the accuracy index of computing resource allocation is less than the accuracy threshold of computing resource allocation, it is determined that the computing resource allocation result of the target task is inaccurate, and the computing resource allocation optimization mechanism of the target task is triggered.

5. The method of claim 4, wherein, The specific analysis process for node matching of the target task is as follows: Determine whether there are nodes in the current node cluster that meet the computing power resource requirements of the target task; If there are nodes in the current node cluster that meet the computing power resource requirements of the target task, then the nodes that meet the computing power resource requirements of the target task are marked as valid nodes. The conflict risk value of each valid node is obtained and sorted in ascending order of conflict risk value. The valid node with the highest conflict risk value is selected for binding to the target task. If there are no nodes in the current node cluster that meet the computing power resource requirements of the target task, the task will be classified and processed based on its priority. If the target task is of high priority, the computing power resource preemption mechanism will be triggered. If the target task is not of high priority, the task will wait for idle computing power resources.

6. The method of claim 1, wherein, The specific determination process for whether to issue an information warning regarding the allocation of computing resources for the target task is as follows: Compare the accurate final value of computing resource allocation with the accurate limit value of computing resource allocation; If the final value of the accurate allocation of computing resources is greater than or equal to the threshold value of the accurate allocation of computing resources, it is determined that no information warning will be issued for the computing resource allocation process of the target task. If the final value of the accurate allocation of computing resources is less than the accurate limit value of the allocation of computing resources, then it is determined that an information warning should be issued for the allocation process of computing resources for the target task.

7. A task-priority-based computing resource allocation system, applying the task-priority-based computing resource allocation method according to any one of claims 1 to 6, characterized in that: include: The target task extraction module is used to extract the multi-dimensional attributes of the target task after the system message bus receives the target task and perform verification processing. The multi-dimensional attributes of the target task that have passed the verification are integrated into a valid dataset. The valid dataset is sent to the priority decision engine to generate an initial priority score and divide the task priority. At the same time, it is determined whether the target task has any unfinished prerequisite dependent tasks. The computing resource allocation module is used to allocate computing resources to target tasks that do not have any unfinished prerequisite tasks, monitor the parameters of the computing resource allocation process, thereby determining whether the computing resource allocation result of the target task is accurate, perform node matching for the target task based on the accurate computing resource allocation result, and trigger the computing resource allocation optimization mechanism for the target task based on the inaccurate computing resource allocation result. The computing resource pre-allocation module is used to pre-allocate computing resources for target tasks with unfinished prerequisite tasks, while acquiring the status monitoring indicators of the target tasks and determining whether the computing resource allocation needs to be adjusted during the execution of the target tasks, thereby realizing computing resource allocation based on task priority.

Images