Computing power distribution and scheduling method and device, electronic equipment and computer program product

Abstract

The invention relates to the field of artificial intelligence, in particular to a computing power distribution and scheduling method and device, electronic equipment and a computer program product. The method comprises the following steps: performing quantification processing on computing power resources of various types of computing power nodes and computing power demands of computing tasks to generate quantification results; determining a target computing power distribution strategy at least based on the quantification result and a preset optimization model, the target computing power distribution strategy being used for indicating to distribute the computing task to a target computing power node with the highest matching degree; and based on the target computing power distribution strategy, distributing and monitoring the computing tasks. By dynamically analyzing and calculating the heterogeneity of task requirements and computing power resources and adopting an optimized allocation strategy and a scheduling mechanism, the resources are effectively utilized, and the task execution efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence, and in particular to a computing power allocation and scheduling method and device, electronic equipment and computer program product. BACKGROUND

[0002] In a heterogeneous computing environment, computing power allocation and scheduling refers to reasonably allocating tasks to the most suitable computing unit according to the requirements of the computing task and the characteristics of the computing unit, and optimizing the use of resources through scheduling strategies to improve overall computing performance and efficiency. At present, the existing computing power allocation and scheduling methods each have their own defects: 1) Allocation and scheduling method based on task characteristic analysis: insufficient accuracy.

[0003] This method often relies on historical data and pre-defined models, but the characteristics of computing tasks will change with the environment and data, resulting in inaccurate task analysis and classification, and thus unreasonable allocation of computing power (for example: tasks are allocated to unsuitable processors), ultimately reducing resource utilization and task execution efficiency.

[0004] 2) Allocation and scheduling method based on dynamic resources: high resource consumption and poor real-time performance.

[0005] This method requires continuous collection and analysis of a large amount of system state and task information, and estimation of future resource requirements, which consumes additional computing resources and prolongs system response time; complex scheduling algorithms introduce delays in high-frequency scheduling and resource reallocation, affecting real-time performance, and the algorithm implementation and maintenance costs are high.

[0006] 3) Allocation and scheduling method based on static algorithms: unbalanced use of heterogeneous resources.

[0007] Different processors have significantly different loads (for example: when a GPU is excessively busy, other processors are idle), especially when task demand fluctuates greatly, resource overuse or waste is a major problem, limiting overall system performance. SUMMARY

[0008] The present disclosure is proposed in view of the above problems. The present disclosure provides a computing power allocation and scheduling method, device, electronic equipment and computer program product.

[0009] According to one aspect of the present disclosure, a computing power allocation and scheduling method is provided, the method comprising: quantitatively processing computing power resources of a plurality of types of computing power nodes and computing power requirements of computing tasks to generate a quantization result; determining a target computing power allocation strategy based at least on the quantization result and a pre-set optimization model, wherein the target computing power allocation strategy is used to indicate allocation of the computing tasks to a target computing power node with the highest matching degree; and allocating the computing tasks and monitoring based on the target computing power allocation strategy.

[0010] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the computing power resources of various types of computing power nodes and the computing power requirements of computing tasks are quantified to generate quantification results, including: quantifying the computing power resources of various types of computing power nodes to generate computing power resource vectors; and quantifying the computing power requirements of computing tasks based on a preset extraction model to generate task requirement vectors.

[0011] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the computing power resources of multiple types of computing power nodes are quantized to generate computing power resource vectors, including: identifying each corresponding resource type among the multiple types of computing power nodes; quantizing the performance parameter vector corresponding to each resource type; quantizing the current load value of each computing power node corresponding to each resource type; and generating a computing power resource vector corresponding to each computing power node based on the resource type, performance parameter vector, and current load value.

[0012] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the computing power requirements of the computing task are quantified based on a preset extraction model to generate a task requirement vector, including: extracting computing power requirement features in the computing task based on the preset extraction model; quantifying the computing power requirement features to generate a task requirement vector.

[0013] Furthermore, according to one aspect of the computing power allocation and scheduling method disclosed herein, the computing power demand characteristics include at least: task type, resource demand vector, and priority value.

[0014] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the computing power demand characteristics are quantified to generate a task demand vector, including: analyzing the task type corresponding to the computing task; quantifying the resource demand vector corresponding to each task type; quantifying the priority value of the computing task; and generating a task demand vector corresponding to the computing task based on the task type, resource demand vector, and priority value.

[0015] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the target computing power allocation strategy is determined based at least on quantization results and a preset optimization model, including: determining the target computing power allocation strategy based on computing power resource vectors, task requirement vectors, preset optimization models, and preset constraints.

[0016] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, determining the target computing power allocation strategy based on computing power resource vector, task demand vector, preset optimization model and preset constraints includes: generating multiple candidate computing power allocation strategies based on computing power resource vector, task demand vector and preset optimization model; and generating the target computing power allocation strategy based on preset optimization model, preset constraints and candidate computing power allocation strategies.

[0017] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the preset constraints include: a computing task is assigned to a computing node for processing; the total load of each computing node is less than or equal to its maximum capacity; and matching is performed based on the task type of the computing task and the performance parameters of the computing node.

[0018] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the allocation and monitoring of computing tasks based on a target computing power allocation strategy includes: allocating computing tasks to target computing power nodes based on the target computing power allocation strategy; monitoring the execution status of computing tasks and / or the working status of target computing power nodes; and performing scheduling when the execution status of computing tasks and / or the working status of target computing power nodes meet a preset threshold.

[0019] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the scheduling includes: generating an updated computing power allocation strategy based on the execution status of computing tasks that meet a preset threshold and / or the working status of the target computing power node, wherein the updated computing power allocation strategy is used to indicate that some or all of the computing tasks are allocated to the updated computing power node; and allocating computing tasks and monitoring them based on the updated computing power allocation strategy.

[0020] Furthermore, according to one aspect of the computing power allocation and scheduling method of this disclosure, the method further includes: training an initial extraction model based on training data to generate a preset extraction model.

[0021] According to another aspect of this disclosure, a computing power allocation and scheduling apparatus is provided, comprising: a quantization processing module for quantifying the computing power resources of various types of computing power nodes and the computing power requirements of computing tasks, and generating quantization results; a strategy generation module for determining a target computing power allocation strategy based at least on the quantization results and a preset optimization model, wherein the target computing power allocation strategy is used to instruct the allocation of computing tasks to the target computing power node with the highest matching degree; and an execution scheduling module for allocating computing tasks and monitoring them based on the target computing power allocation strategy.

[0022] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the quantization processing module includes: a resource quantization submodule, used to quantify the computing power resources of various types of computing power nodes to generate computing power resource vectors; and a demand quantization submodule, used to quantify the computing power demand of computing tasks based on a preset extraction model to generate task demand vectors.

[0023] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the resource quantization submodule includes: a type identification unit for identifying each corresponding resource type among multiple types of computing power nodes; a performance quantization unit for quantizing the performance parameter vector corresponding to each resource type; a load quantization unit for quantizing the current load value of each computing power node corresponding to each resource type; and a resource integration unit for generating a computing power resource vector corresponding to each computing power node based on the resource type, performance parameter vector, and current load value.

[0024] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the demand quantization submodule includes: a feature extraction unit, used to extract computing power demand features in the computing task based on a preset extraction model; and a feature quantization unit, used to quantify the computing power demand features to generate a task demand vector.

[0025] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the computing power demand characteristics include at least: task type, resource demand vector, and priority value.

[0026] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the feature quantization unit includes: a type analysis subunit for analyzing the task type corresponding to the computing task; a demand quantization subunit for quantifying the resource demand vector corresponding to each task type; a priority quantization subunit for quantifying the priority value of the computing task; and a demand integration subunit for generating the task demand vector corresponding to the computing task based on the task type, resource demand vector, and priority value.

[0027] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the strategy generation module includes: a strategy generation unit, used to determine a target computing power allocation strategy based on a computing power resource vector, a task requirement vector, a preset optimization model, and preset constraints.

[0028] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the strategy generation unit includes: a candidate strategy subunit, used to generate multiple candidate computing power allocation strategies based on computing power resource vectors, task requirement vectors and preset optimization models; and a target strategy subunit, used to generate a target computing power allocation strategy based on preset optimization models, preset constraints and candidate computing power allocation strategies.

[0029] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the preset constraints include: a computing task is assigned to a computing node for processing; the total load of each computing node is less than or equal to its maximum capacity; and matching is performed based on the task type of the computing task and the performance parameters of the computing node.

[0030] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the execution scheduling module includes: a task allocation unit, used to allocate computing tasks to target computing power nodes based on a target computing power allocation strategy; and a monitoring scheduling unit, used to monitor the execution status of computing tasks and / or the working status of target computing power nodes, and to execute scheduling when the execution status of computing tasks and / or the working status of target computing power nodes meet a preset threshold.

[0031] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the monitoring and scheduling unit includes: an update strategy subunit, used to generate an update computing power allocation strategy based on the execution status of computing tasks that meet preset thresholds and / or the working status of target computing power nodes, wherein the update computing power allocation strategy is used to indicate that some or all of the computing tasks are allocated to the update computing power nodes; and an update allocation subunit, used to allocate computing tasks and monitor them based on the update computing power allocation strategy.

[0032] Furthermore, according to one aspect of the computing power allocation and scheduling apparatus of this disclosure, the apparatus further includes: a model training module, used to train an initial extraction model based on training data to generate a preset extraction model.

[0033] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory for storing computer-readable instructions; and a processor for executing the computer-readable instructions, causing the electronic device to perform the computing power allocation and scheduling method as described above.

[0034] According to another aspect of this disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, it implements the computing power allocation and scheduling method described above.

[0035] As will be described in detail below, the computing power allocation and scheduling method according to embodiments of this disclosure dynamically analyzes the computing task requirements and the heterogeneity of computing resources, and adopts optimized allocation strategies and scheduling mechanisms to achieve effective resource utilization and improve task execution efficiency. Specifically, by accurately quantifying the information of heterogeneous computing resources, a reliable basis is provided for task analysis and allocation decisions; by deeply extracting and accurately quantifying the characteristics of computing tasks, accurate analysis of task requirements is ensured; by formulating optimal computing power allocation strategies, resource utilization efficiency and task execution efficiency are improved; and by monitoring task execution and resource usage in real time according to the scheduling mechanism during task execution based on the computing power allocation strategy, and further adjusting strategies and algorithms based on feedback data, overall continuous optimization is achieved.

[0036] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0037] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0038] Figure 1 This is a schematic diagram illustrating an application scenario of the computing power allocation and scheduling method according to an embodiment of the present disclosure.

[0039] Figure 2 This is a flowchart illustrating a computing power allocation and scheduling method according to an embodiment of the present disclosure.

[0040] Figure 3 This is a flowchart illustrating the overall process of computing power allocation and scheduling method according to an embodiment of the present disclosure.

[0041] Figure 4 This is a schematic diagram of a computing power allocation and scheduling device according to an embodiment of the present disclosure.

[0042] Figure 5 This is a hardware block diagram illustrating an electronic device according to an embodiment of the present disclosure.

[0043] Figure 6 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0045] First, refer to Figure 1 Overview of application scenarios according to embodiments of this disclosure.

[0046] Figure 1 This is a schematic diagram illustrating an application scenario of the computing power allocation and scheduling method according to an embodiment of this disclosure. For example... Figure 1 As shown, the application scenario includes at least: a central node 11 and multiple computing nodes 12 (e.g., computing node 1, computing node 2, ... computing node n). The computing nodes 12 are nodes responsible for executing the computing tasks assigned by the central node 11 (i.e., computing power allocation and scheduling device 11). They include various types to form a heterogeneous computing environment.

[0047] In one embodiment of this disclosure, the computing node 12 may include different types of processors, such as a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), a Neural Processing Unit (NPU), and a Machine Learning Processing Unit (MLU), without specific limitations. These types of processors each have their own computational advantages and can be used to handle different types of computing tasks. Specifically: 1) CPU: It excels in logic control, multi-task concurrent scheduling and execution of complex instruction sequences. It is the core carrier of general computing tasks and has significant competitiveness in handling business scenarios with high real-time responsiveness and low latency.

[0048] 2) GPU: It has the advantages of large-scale parallel computing architecture. It was originally designed for graphics and image rendering needs, and has since been widely used in parallel-intensive tasks such as deep learning training and inference, and scientific numerical computing. It can efficiently support the parallel operation of massive amounts of data.

[0049] 3) NPU: Deeply optimized for neural network operations in the field of artificial intelligence, it can efficiently process core operators such as convolution, pooling, and activation functions, and has the characteristics of low power consumption and high computing power density, making it suitable for artificial intelligence (AI) exclusive tasks such as deep learning inference and intelligent image processing.

[0050] 4) MLU: Customized optimization for machine learning algorithms, with efficient processing capabilities for core machine learning operations such as matrix operations and data iterative training, which can significantly improve the execution efficiency of machine learning tasks.

[0051] As mentioned above, to address the issues of insufficient accuracy, high complexity, and uneven resource utilization in existing technologies, this disclosure proposes a computing power allocation and scheduling method, for details of which can be found in [link to relevant documentation]. Figures 2-3 .

[0052] Figure 2 This is a flowchart illustrating a computing power allocation and scheduling method according to an embodiment of the present disclosure. Figure 2 As shown, the computing power allocation and scheduling method applied to the central node 11 according to this embodiment of the disclosure may include at least the following steps.

[0053] In step S201, the computing resources and computing requirements of various types of computing nodes are quantified to generate quantification results. As mentioned above, computing nodes 12 include various types, and different types of computing nodes 12 are suitable for handling different computing tasks. To facilitate subsequent matching of resources and requirements, this step standardizes computing resources and computing requirements, aiming to establish a quantifiable, computable, and comparable global computing resource profile and computing task profile, providing data support for subsequent steps.

[0054] Understandably, this step, by constructing a standardized representation of computing resources, replaces the high-frequency full-volume collection mode of global computing resources in existing technologies, significantly reducing the additional resource overhead generated in the information collection and analysis process. At the same time, the preset feature extraction model used in the task computing power requirement quantification process has the ability to adapt to the dynamic changes of task characteristics with the environment and data, and can achieve a precise standardized description of computing tasks, providing a data foundation for subsequent task-resource matching.

[0055] In step S202, a target computing power allocation strategy is determined based at least on the quantization results and a preset optimization model. This strategy instructs the allocation of computing tasks to the target computing power node with the highest matching degree. In other words, this step, based on the quantization of computing resources and computing power requirements in step S201, combined with the dynamic algorithm in the preset optimization model, solves the problems of uneven resource utilization and insufficient accuracy in existing technologies through multiple rounds of iteration to find the optimal solution. This achieves precise matching of computing resources and computing tasks, effectively improving the overall operating efficiency and resource utilization in the computing environment.

[0056] In step S203, computing tasks are allocated and monitored based on the target computing power allocation strategy. That is, this step executes the target computing power allocation strategy from step S202, distributes computing tasks to the target computing power nodes, and continuously monitors the execution of these tasks. This ensures timely scheduling in case of anomalies or changes on the computing power resource side and / or the computing task side, fundamentally addressing the issue of insufficient allocation accuracy and improving allocation rationality and overall operational stability. The entire process will be... Figure 3 Further explanation will follow.

[0057] Figure 3 This is a flowchart illustrating the overall process of heterogeneous computing power allocation and scheduling method according to an embodiment of this disclosure. Figure 3 As shown, the heterogeneous computing power allocation and scheduling method applied to the central node 11 in this embodiment of the present disclosure may include the following steps.

[0058] Step S1: Quantify the computing resources of various types of computing nodes to generate computing resource vectors. In other words, this step aims to unify the description rules of computing resources across different computing nodes, transforming the computing power and availability of all types of computing nodes into standardized data, constructing a global computing resource profile, and providing data support for subsequent task analysis and matching.

[0059] In one embodiment of this disclosure, the computing resources of the computing unit may include: resource type, performance parameter vector, and current load value. Specifically: S11. Identify the corresponding resource type for each of the various types of computing power nodes.

[0060] The central node identifies the type of all computing power nodes (or processors), including CPU, GPU, NPU, MLU, etc., and then quantifies them uniformly to form a resource type set. ,

[0061] in, This represents the i-th type of processor. In one example embodiment of this disclosure, It can represent the CPU. It can represent a GPU, and so on.

[0062] S12. Quantize the performance parameter vector corresponding to each of the resource types.

[0063] Different types of processors have different performance parameters, which can be used to describe the computing power and characteristics of the processor. This step generates a performance parameter vector by quantizing the performance parameters to standardize the computing power of the computing nodes. ,

[0064] in, This represents the j-th performance parameter vector of the i-th processor, where each performance parameter vector includes multiple performance parameters.

[0065] In one embodiment of this disclosure, the performance parameters of a CPU processor may include clock frequency, number of cores, cache size, power consumption, etc., and its performance parameter vector can be represented as:

[0066] Similarly, for GPU processors, performance parameters may include computing power (such as floating-point operations per second (FLOPS)), memory size, bandwidth, and the number of CUDA cores. For NPU processors, performance parameters may include neural network processing capabilities, power consumption, on-chip storage, etc. For an MLU processor, performance parameters may include machine learning acceleration capabilities, bandwidth, on-chip storage, etc.

[0067] S13. Quantify the current load value of each computing node corresponding to each resource type. That is, in addition to quantifying the computing power in step S12, availability also needs to be quantified to objectively describe the current state of each computing node. This step aims to quantify the current load to standardize the availability of computing nodes and generate the current load matrix L.

[0068] in, This represents the current load value of the j-th instance of the i-th processor.

[0069] It is understandable that the dimension of the load matrix L is n×max( ), where n represents the total number of processor types, This represents the number of instances of the i-th type of processor, that is, max( The maximum value among all processor types is taken to avoid data structure chaos caused by different numbers of processor instances of different types.

[0070] S14. Based on the resource type, the performance parameter vector, and the current load value, generate the computing power resource vector corresponding to each computing power node. In other words, this step integrates the various parameters (resource type, performance parameter vector, current load value) used to describe computing power resources in steps S11-S13 above, aiming to construct a global computing power resource profile, which can be represented by the computing power resource vector. express,

[0071] in, This represents the computing resource vector of the k-th processor instance. This indicates the type of processor it belongs to. This represents the performance parameter vector corresponding to its type. This represents its current load value (i.e., the current load value of the jth instance of the i-th processor).

[0072] In summary, the standardized mathematical representation methods described above enable the accurate abstraction and quantification of heterogeneous computing resources, providing standardized input data for subsequent allocation strategies. Furthermore, these methods allow for the real-time collection and updating of computing node status parameters, ensuring that the central node can dynamically perceive changes in the global resource status. This drives adaptive optimization of subsequent computing power allocation strategies, improving overall computational efficiency.

[0073] Step S2: Based on the preset extraction model, the computing power requirements of the computing task are quantified to generate a task requirement vector. In other words, this step aims to accurately transform the task description into computing power requirement features, and then transform these features into standardized data to construct a computing task profile, providing data support for subsequent task analysis and matching. Specifically: S21. Based on the preset extraction model, extract the computing power requirement features in the computing task.

[0074] Specifically, based on training data, the initial extraction model is trained to generate a preset extraction model. That is, the preset extraction model can be pre-trained, and its training data can include previously completed computational tasks and / or simulated datasets. This training data includes at least the descriptive information of the computational tasks and the execution results of those tasks. Through training, the model can learn the feature differences between different task types and the corresponding optimal feature extraction methods, thereby improving the mapping between task descriptions and resource requirements, ultimately resulting in a convergent and highly accurate preset extraction model.

[0075] In one embodiment of this disclosure, the computing power requirement characteristics may include at least: task type, resource requirement vector, and priority value.

[0076] S22. Quantify the computing power requirement features to generate the task requirement vector. That is, after the preset extraction model is trained, it is deployed to extract the aforementioned computing power requirement features based on the description information of the computing task, facilitating subsequent steps. Specifically: (1) Analyze the task type corresponding to the calculation task.

[0077] The central node extracts all task types from the list of tasks to be executed, forming a task type set. ,

[0078] in, This represents the i-th type of task. In one example embodiment of this disclosure, It can represent image processing. It can represent deep learning training, It can represent scientific computing... and so on.

[0079] (2) Quantify the resource requirement vector corresponding to each task type.

[0080] Different types of tasks have different resource requirements. This step quantifies the resource requirements to standardize the processor requirements for each task type, generating a resource requirement vector. ,

[0081] in, This represents the resource requirements of the i-th task on the j-th processor.

[0082] In one example embodiment of this disclosure,

[0083] in, This represents the total resource requirement of the i-th task for CPU instances. This represents the total resource requirement of the GPU instance for the i-th task. This represents the total resource requirement of the NPU instance for the i-th task. This represents the total resource requirement of the i-th task for the MLU instance.

[0084] (3) Quantify the priority value of the calculation task.

[0085] The priority value of a task is used to represent its importance and urgency, and can be expressed as a scalar. It means that among them represents the priority of the i-th task; a higher value indicates a higher priority.

[0086] In one embodiment of this disclosure, the priority can be determined based on factors such as the business importance of the computing task and the deadline, wherein, It can be calculated using the following formula:

[0087] in, and These are the weighting coefficients. `current_time` represents the deadline for the i-th type of task, and `current_time` represents the current time. This represents the business importance of the i-th type of task.

[0088] (4) Based on the task type, resource requirement vector, and priority value, generate the task requirement vector corresponding to the computing task. That is to say, this step integrates the various features (task type, resource requirement vector, priority value) used to describe computing power requirements in steps (1)-(3) above, aiming to construct a computing task profile, which can be represented by the task requirement vector. express,

[0089] in, The task requirement vector representing the i-th type of task. Represents the i-th type of task. This represents the resource requirements for the i-th type of task. This represents the priority of the i-th task.

[0090] In summary, the aforementioned feature extraction and standardized mathematical representation methods can accurately extract and quantify computational task requirements, providing standardized input data for subsequent allocation strategies. Furthermore, these methods can be used to adapt to dynamic changes in task requirements in real time, driving adaptive optimization of subsequent computing power allocation strategies and improving overall computational efficiency.

[0091] Step S3: Based on the computing power resource vector, task requirement vector, preset optimization model, and preset constraints, determine the target computing power allocation strategy. As mentioned above, the target computing power allocation strategy is used to indicate which computing tasks are assigned to the target computing power node with the highest matching degree. In other words, this step aims to formulate the optimal "task-resource" mapping based on the characteristics of heterogeneous computing power resources and task requirement features, to ensure that each task can be processed on the most suitable processor, thereby maximizing overall processing performance. Specifically: Step S31: Based on the computing power resource vector, task requirement vector and preset optimization model, generate multiple candidate computing power allocation strategies.

[0092] Specifically, the computing power resource vector generated in step S1 and the task requirement vector generated in step S2 As a data input preset optimization model, the preset optimization model first randomly generates a batch of "task-resource" mapping relationships based on its built-in algorithm as candidate computing power allocation strategies.

[0093] Step S32: Generate the target computing power allocation strategy based on the preset optimization model, preset constraints and candidate computing power allocation strategies.

[0094] Specifically, firstly, after generating a batch of candidate computing power allocation strategies in step S31, the preset optimization model scores each candidate computing power allocation strategy according to its built-in objective function F.

[0095] In one embodiment of this disclosure, the objective function F can be described by the following formula:

[0096] in, Representative task priority, Representative task In resources Resource utilization rate of the above execution Representative task Allocated to resources The decision variable takes the value 0 or 1 (0 represents not assigning the task). Allocate to resources 1 represents the task Allocate to resources ), m represents the number of resources, and n represents the number of tasks.

[0097] In other words, the objective function F aims to maximize resource utilization efficiency and task execution efficiency. Specifically, the objective function F combines resource utilization and task priority to balance system performance and task urgency.

[0098] in, By comprehensively considering resources Performance parameters ,Task resource requirements and current load It is certain that its calculation formula can be expressed as:

[0099] Wherein, function f represents resource requirement. and performance parameters The degree of matching between them can be determined using the following formula:

[0100] in, Represents the weighting coefficient. This represents the performance requirement on the k-th resource. The performance parameters on the k-th resource represent the importance of different resource requirements in task execution. Current load. Representative Resources The current load status is used to prevent overload and to consider load balancing.

[0101] Furthermore, pre-set constraints are also provided to ensure the rationality of task allocation and the effectiveness of resource utilization. In one embodiment of this disclosure, the pre-set constraints may include: <1> A computing task is assigned to a computing node for processing.

[0102] In other words, the uniqueness of task allocation is constrained, meaning each task can only be assigned to one processor instance for processing. This can be expressed by the formula:

[0103] <2> The total load of each computing node is less than or equal to its maximum capacity.

[0104] In other words, the total load on each processor instance cannot exceed its maximum capacity, which can be expressed by the formula:

[0105] in, Indicates task The resource demand vector, For resources The capacity vector. According to the performance parameters provided by the resource description module It is calculated. For example, for a CPU, capacity may include the number of available cores and maximum processing power; for a GPU, capacity may include video memory size and computing power.

[0106] <3> Matching is performed based on the task type of the computing task and the performance parameters of the computing nodes.

[0107] In other words, tasks can only be assigned to processors that are suitable for their type, which can be expressed by the formula:

[0108] in, For the task type For resources Supported task types.

[0109] Secondly, based on the above scores, updates and iterations are performed. That is, the higher the score, the closer it is to the optimal mapping. Candidate computing power allocation strategies with high scores are retained, while those with low scores are eliminated. Then, crossover, mutation, and other operations can be performed on the retained candidate computing power allocation strategies. Through multiple rounds of iteration, the algorithm is solved until it converges.

[0110] Finally, the candidate computing power allocation strategy with the highest score after convergence is selected as the target computing power allocation strategy.

[0111] In one embodiment of this disclosure, the target computing power allocation strategy may include: task identifier, allocated processor instance (i.e. target computing power node), and expected resource usage.

[0112] In one embodiment of this disclosure, in order to improve the solution efficiency and adapt to dynamic environments, the preset optimization model can embed a genetic algorithm, which can find a near-optimal solution in a short time to meet real-time requirements.

[0113] In summary, this step, based on the global computing power resource profile and computing task profile, determines the optimal solution (i.e., the target computing power allocation strategy) for the "task-resource" mapping that takes into account task priority, resource utilization, and load balancing by using a preset optimization model and preset constraints. This target computing power allocation strategy can allocate computing tasks to the target computing power node with the highest matching degree (i.e., ensure that each task can be processed on the most suitable processor), thereby achieving global optimal scheduling and maximizing overall processing performance.

[0114] Step S4: Based on the target computing power allocation strategy, allocate computing tasks to the target computing power nodes.

[0115] Specifically, the optimal mapping strategy of "task-resource" formulated in step S3 is analyzed to generate a specific task plan, and then the plan is executed to allocate the task to the most suitable target computing power node to ensure the task is implemented and executed.

[0116] Specifically, first examine the target resources. The current state of the target computing node (i.e., the current state of the target computing node) is checked to ensure that it has sufficient available resources to start the task. If resources are sufficient, the task will be... Hand over resources It then updates the current load information of the resources.

[0117] Step S5: Monitor the execution status of the computing task and / or the working status of the target computing node. When the execution status of the computing task and / or the working status of the target computing node meet the preset threshold, execute the scheduling.

[0118] In other words, the usage of various resources is continuously monitored during task execution. Specifically, the load status of each computing node is monitored in real time to promptly detect changes in resource utilization and track the task execution status. In one embodiment of this disclosure, the monitored information may include processor utilization, memory usage, task progress, etc.

[0119] Furthermore, when significant changes in resource usage or abnormalities in task execution are detected (i.e., when preset thresholds are met), the task scheduling plan can be adjusted in real time according to the scheduling mechanism.

[0120] Specifically, in one embodiment of this disclosure, the scheduling mechanism includes: generating an updated computing power allocation strategy based on the execution status of computing tasks that meet a preset threshold and / or the working status of the target computing power node, wherein the updated computing power allocation strategy is used to indicate that some or all of the computing tasks are allocated to the updated computing power node; and allocating computing tasks and monitoring them based on the updated computing power allocation strategy.

[0121] In other words, this mechanism first collects various types of data in real time, including task execution progress, processor utilization, and memory and storage usage. Then, it processes the collected data, continuously evaluating the difference between the actual execution performance and the planned performance. If bottlenecks or uneven resource utilization are detected during task execution, feedback data is immediately generated, and an updated computing power allocation strategy is developed based on this feedback data. Furthermore, the task scheduling plan is adjusted in real time according to the updated computing power allocation strategy to ensure that tasks can continue to execute efficiently on the new resources. For example, when the load on a processor instance is too high, some tasks can be migrated to processors with lower loads according to the updated computing power allocation strategy to balance the use of system resources.

[0122] Understandably, through this scheduling mechanism, central node 11 can adaptively optimize resource allocation and task scheduling, improving overall computing efficiency and resource utilization. This mechanism ensures that high performance and high stability are maintained in complex and ever-changing heterogeneous computing environments.

[0123] It should be noted that the central node 11 can also perform scheduling management based on task priority to ensure that high-priority tasks are processed first. In one embodiment of this disclosure, the central node 11 can reserve more resources or adopt a preemptive scheduling strategy for high-priority tasks to ensure their timely completion.

[0124] When the task Once completed, remove it from the resources. Remove the resource and release the occupied resources. The central node 11 updates the resource load information and prepares to accept new task scheduling requests.

[0125] In summary, steps 4 and S5 ensure that computing tasks can be efficiently allocated and executed, improving overall performance and resource utilization. This not only optimizes computing power allocation strategies but also provides real-time response and dynamic adjustment capabilities, adapting to the complexity of heterogeneous computing environments and the diversity of task requirements.

[0126] Figure 4 This is a schematic diagram of a computing power allocation and scheduling apparatus according to an embodiment of the present disclosure. Figure 4 As shown, the computing power allocation and scheduling device 11 may include at least the following modules.

[0127] The quantization processing module 111 is used to quantify the computing resources and computing requirements of various types of computing power nodes and computing tasks, and generate quantization results. The strategy generation module 112 is used to determine the target computing power allocation strategy based at least on the quantization results and the preset optimization model, wherein the target computing power allocation strategy is used to indicate that computing tasks are allocated to the target computing power node with the highest matching degree. The execution scheduling module 113 allocates computing tasks and monitors them based on the target computing power allocation strategy.

[0128] Furthermore, the quantization processing module 111 includes: The resource quantization submodule 1111 is used to quantify the computing resources of various types of computing power nodes and generate computing resource vectors. The demand quantification submodule 1112 is used to quantify the computing power requirements of computing tasks based on a preset extraction model and generate a task demand vector.

[0129] The resource quantification submodule 1111 may further include: Type identification unit 11111 is used to identify each corresponding resource type among multiple types of computing power nodes; The performance quantization unit 11112 is used to quantize the performance parameter vector corresponding to each resource type; The load quantization unit 11113 is used to quantify the current load value of each computing node corresponding to each resource type; The resource integration unit 11114 is used to generate a computing power resource vector corresponding to each computing power node based on the resource type, performance parameter vector, and current load value.

[0130] The demand quantification submodule 1112 may further include: The feature extraction unit 11121 is used to extract computing power demand features in computing tasks based on a preset extraction model. The feature quantization unit 11122 is used to quantize the computing power requirement features and generate a task requirement vector.

[0131] Among them, the computing power demand characteristics include at least: task type, resource demand vector, and priority value.

[0132] The feature quantization unit 11122 may further include: Type analysis subunit 111221 is used to analyze the task type corresponding to the computation task; Demand quantification subunit 111222 is used to quantify the resource demand vector corresponding to each task type; Priority quantization subunit 111223 is used to quantize the priority value of the calculation task; The requirement integration subunit 111224 is used to generate the task requirement vector corresponding to the computing task based on the task type, resource requirement vector, and priority value.

[0133] Additionally, the policy generation module 112 may include: The strategy generation unit 1121 is used to determine the target computing power allocation strategy based on the computing power resource vector, task requirement vector, preset optimization model and preset constraints.

[0134] Furthermore, the strategy generation unit 1121 includes: The candidate strategy subunit 11211 is used to generate multiple candidate computing power allocation strategies based on the computing power resource vector, task requirement vector and preset optimization model; The target strategy subunit 11212 is used to generate a target computing power allocation strategy based on a preset optimization model, preset constraints and candidate computing power allocation strategies.

[0135] The preset constraints include: a computing task is assigned to a computing node for processing; the total load of each computing node is less than or equal to its maximum capacity; and matching is performed based on the task type of the computing task and the performance parameters of the computing node.

[0136] Additionally, the execution scheduling module 113 may include: The task allocation unit 1131 is used to allocate computing tasks to target computing nodes based on the target computing power allocation strategy; The monitoring and scheduling unit 1132 is used to monitor the execution status of computing tasks and / or the working status of target computing nodes. When the execution status of computing tasks and / or the working status of target computing nodes meet the preset threshold, scheduling is performed.

[0137] The monitoring and scheduling unit 1132 may further include: The update strategy subunit 11321 is used to generate an update computing power allocation strategy based on the execution status of computing tasks that meet a preset threshold and / or the working status of the target computing power node. The update computing power allocation strategy is used to indicate that part or all of the computing tasks are allocated to the update computing power node. The update allocation subunit 11322 is used to allocate computing tasks and monitor them based on the updated computing power allocation strategy.

[0138] In addition, the computing power allocation and scheduling device 11 may also include: The model training module 114 is used to train the initial extraction model based on the training data to generate a preset extraction model.

[0139] In summary, the resource quantization submodule 1111 in the quantization processing module 111 can accurately quantify information about heterogeneous computing resources, providing a reliable foundation for task analysis and allocation decisions; the demand quantization submodule 1112 in the quantization processing module 111 can deeply extract the characteristics of computing tasks, ensuring accurate analysis of task requirements; the strategy generation module 112, based on the data from the resource quantization submodule 1111 and the demand quantization submodule 1112, formulates the optimal computing power allocation strategy to improve resource utilization efficiency and task execution efficiency; the execution scheduling module 113, on the one hand, maps and schedules tasks and resources according to the computing power allocation strategy to ensure timely task completion, and on the other hand, monitors task execution and resource usage in real time according to the scheduling mechanism, adjusts strategies and algorithms based on feedback data, and achieves continuous optimization of the whole. This comprehensive design and real-time optimization capability of the computing power allocation and scheduling device 11 effectively solves the problems of insufficient accuracy, high complexity, and uneven resource utilization in existing technologies.

[0140] Figure 5 This is a hardware block diagram illustrating an electronic device according to an embodiment of the present disclosure. The electronic device according to an embodiment of the present disclosure includes at least a processor and a memory for storing computer-readable instructions. When the computer-readable instructions are loaded and executed by the processor, the processor performs the computing power allocation and scheduling method as described above.

[0141] Figure 5 The illustrated electronic device 500 specifically includes a central processing unit (CPU) 501, a graphics processing unit (GPU) 502, and a memory 503. These units are interconnected via a bus 504. The CPU 501 and / or GPU 502 can function as the aforementioned processor, and the main memory 503 can function as the aforementioned memory storing computer-readable instructions. Furthermore, the electronic device 500 may also include a communication unit 505, a storage unit 506, an output unit 507, an input unit 508, and an external device 509, all of which are also connected to the bus 504.

[0142] Figure 6 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. Figure 6 As shown, a computer program product 600 according to an embodiment of this disclosure stores a computer program 601. When the computer program 601 is executed by a processor, it performs the computing power allocation and scheduling method described with reference to the above figures. The computer program product includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0143] The above description, with reference to the accompanying drawings, outlines a computing power allocation and scheduling method, apparatus, electronic device, and computer program product according to embodiments of the present disclosure. The computing power allocation and scheduling method according to embodiments of the present disclosure dynamically analyzes the heterogeneity of computing task requirements and computing resources, employing optimized allocation strategies and scheduling mechanisms to achieve effective resource utilization and improve task execution efficiency. Specifically, it provides a reliable foundation for task analysis and allocation decisions by accurately quantifying information about heterogeneous computing resources; it ensures accurate analysis of task requirements by deeply extracting and accurately quantifying the characteristics of computing tasks; it improves resource utilization efficiency and task execution efficiency by formulating optimal computing power allocation strategies; and it achieves continuous overall optimization by monitoring task execution and resource usage in real time according to the scheduling mechanism during task execution based on the computing power allocation strategy and further adjusting strategies and algorithms based on feedback data.

[0144] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0145] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0146] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0147] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0148] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0149] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0150] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0151] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for allocating and scheduling computing power, characterized in that, The method includes: The computing resources and computing requirements of various types of computing nodes are quantified to generate quantification results. Based at least on the quantization results and the preset optimization model, a target computing power allocation strategy is determined, wherein the target computing power allocation strategy is used to indicate that the computing task is allocated to the target computing power node with the highest matching degree. Based on the target computing power allocation strategy, the computing tasks are allocated and monitored.

2. The computing power allocation and scheduling method as described in claim 1, characterized in that, The process of quantifying the computing resources and computing requirements of various types of computing nodes and computing tasks to generate quantification results includes: The computing resources of the various types of computing nodes are quantified to generate computing resource vectors; Based on a pre-defined extraction model, the computing power requirements of the computing task are quantified to generate a task requirement vector.

3. The computing power allocation and scheduling method as described in claim 2, characterized in that, The process of quantifying the computing resources of various types of computing nodes to generate computing resource vectors includes: Identify the corresponding resource type for each of the various types of computing power nodes; Quantize the performance parameter vector corresponding to each of the resource types; Quantify the current load value of each computing node corresponding to each of the resource types; Based on the resource type, the performance parameter vector, and the current load value, a computing power resource vector corresponding to each computing power node is generated.

4. The computing power allocation and scheduling method as described in claim 2, characterized in that, The step of quantifying the computing power requirements of the computational task based on a preset extraction model to generate a task requirement vector includes: Based on the preset extraction model, extract the computing power requirement features in the computing task; The computing power requirement characteristics are quantified to generate the task requirement vector.

5. The computing power allocation and scheduling method as described in claim 4, characterized in that, The computing power requirement characteristics include at least the following: Task type, resource requirement vector, priority value.

6. The computing power allocation and scheduling method as described in claim 5, characterized in that, The step of quantifying the computing power demand characteristics to generate the task demand vector includes: Analyze the task type corresponding to the computational task; Quantize the resource requirement vector corresponding to each of the aforementioned task types; Quantify the priority value of the computation task; Based on the task type, the resource requirement vector, and the priority value, the task requirement vector corresponding to the computing task is generated.

7. The computing power allocation and scheduling method as described in claim 2, characterized in that, The determination of the target computing power allocation strategy, based at least on the quantization results and the preset optimization model, includes: Based on the computing power resource vector, the task requirement vector, the preset optimization model, and the preset constraints, the target computing power allocation strategy is determined.

8. The computing power allocation and scheduling method as described in claim 7, characterized in that, The step of determining the target computing power allocation strategy based on the computing power resource vector, the task requirement vector, the preset optimization model, and preset constraints includes: Based on the computing power resource vector, the task requirement vector, and the preset optimization model, multiple candidate computing power allocation strategies are generated. Based on the preset optimization model, the preset constraints, and the candidate computing power allocation strategy, the target computing power allocation strategy is generated.

9. The computing power allocation and scheduling method as described in claim 7, characterized in that, The preset constraints include: One of the computing tasks is assigned to one of the computing nodes for processing; The total load of each computing node is less than or equal to its maximum capacity; The matching is based on the task type of the computing task and the performance parameters of the computing node.

10. The computing power allocation and scheduling method as described in claim 1, characterized in that, The allocation and monitoring of computing tasks based on the target computing power allocation strategy includes: Based on the target computing power allocation strategy, the computing tasks are allocated to the target computing power nodes; Monitor the execution status of the computing task and / or the working status of the target computing node, and execute scheduling when the execution status of the computing task and / or the working status of the target computing node meets a preset threshold.

11. The computing power allocation and scheduling method as described in claim 10, characterized in that, The scheduling includes: Based on the execution status of the computing task that meets the preset threshold and / or the working status of the target computing node, an updated computing power allocation strategy is generated, wherein the updated computing power allocation strategy is used to indicate that part or all of the computing task is allocated to the updated computing power node. Based on the updated computing power allocation strategy, the computing tasks are allocated and monitored.

12. The computing power allocation and scheduling method as described in claim 4, characterized in that, The method further includes: Based on the training data, the initial extraction model is trained to generate the preset extraction model.

13. A computing power allocation and scheduling device, characterized in that, The device includes: The quantization processing module is used to quantify the computing resources and computing requirements of various types of computing power nodes and computing tasks, and generate quantization results. The strategy generation module is used to determine a target computing power allocation strategy based at least on the quantization results and a preset optimization model, wherein the target computing power allocation strategy is used to indicate that the computing task is allocated to the target computing power node with the highest matching degree. The execution scheduling module allocates the computing tasks based on the target computing power allocation strategy and monitors them.

14. The computing power allocation and scheduling device as described in claim 13, characterized in that, The quantization processing module includes: The resource quantization submodule is used to quantify the computing resources of the various types of computing power nodes and generate computing resource vectors. The demand quantification submodule is used to quantify the computing power requirements of computing tasks based on a preset extraction model, and generate a task demand vector.

15. The computing power allocation and scheduling device as described in claim 14, characterized in that, The resource quantification submodule includes: A type identification unit is used to identify the corresponding resource type of each of the multiple types of computing power nodes; A performance quantization unit is used to quantize the performance parameter vector corresponding to each of the resource types. A load quantization unit is used to quantify the current load value of each computing node corresponding to each resource type; The resource integration unit is used to generate the computing power resource vector corresponding to each computing power node based on the resource type, the performance parameter vector, and the current load value.

16. The computing power allocation and scheduling device as described in claim 14, characterized in that, The demand quantification submodule includes: The feature extraction unit is used to extract the computing power requirement features in the computing task based on the preset extraction model. The feature quantization unit is used to quantize the computing power requirement features and generate the task requirement vector.

17. The computing power allocation and scheduling device as described in claim 16, characterized in that, The computing power requirement characteristics include at least the following: Task type, resource requirement vector, priority value.

18. The computing power allocation and scheduling device as described in claim 17, characterized in that, The feature quantization unit includes: The type analysis subunit is used to analyze the task type corresponding to the computing task. The demand quantification subunit is used to quantify the resource demand vector corresponding to each of the task types. A priority quantization subunit is used to quantize the priority value of the computation task; The demand integration subunit is used to generate the task demand vector corresponding to the computing task based on the task type, the resource demand vector, and the priority value.

19. The computing power allocation and scheduling device as described in claim 14, characterized in that, The strategy generation module includes: The strategy generation unit is used to determine the target computing power allocation strategy based on the computing power resource vector, the task requirement vector, the preset optimization model, and preset constraints.

20. The computing power allocation and scheduling device as described in claim 19, characterized in that, The strategy generation unit includes: The candidate strategy subunit is used to generate multiple candidate computing power allocation strategies based on the computing power resource vector, the task requirement vector, and the preset optimization model. The target strategy subunit is used to generate the target computing power allocation strategy based on the preset optimization model, the preset constraints, and the candidate computing power allocation strategy.

21. The computing power allocation and scheduling device as described in claim 19, characterized in that, The preset constraints include: One of the computing tasks is assigned to one of the computing nodes for processing; The total load of each computing node is less than or equal to its maximum capacity; The matching is based on the task type of the computing task and the performance parameters of the computing node.

22. The computing power allocation and scheduling device as described in claim 13, characterized in that, The execution scheduling module includes: The task allocation unit is used to allocate the computing tasks to the target computing power nodes based on the target computing power allocation strategy. The monitoring and scheduling unit is used to monitor the execution status of the computing task and / or the working status of the target computing node. When the execution status of the computing task and / or the working status of the target computing node meet a preset threshold, scheduling is performed.

23. The computing power allocation and scheduling device as described in claim 22, characterized in that, The monitoring and scheduling unit includes: An update strategy subunit is used to generate an update computing power allocation strategy based on the execution status of the computing task and / or the working status of the target computing power node that meet the preset threshold, wherein the update computing power allocation strategy is used to indicate that part or all of the computing task is allocated to the update computing power node. The update allocation subunit is used to allocate the computing tasks and monitor them based on the updated computing power allocation strategy.

24. The computing power allocation and scheduling device as described in claim 16, characterized in that, The device further includes: The model training module is used to train the initial extraction model based on the training data to generate the preset extraction model.

25. An electronic device, characterized in that, include: Memory, used to store computer-readable instructions; as well as A processor for executing the computer-readable instructions, causing the electronic device to perform the computing power allocation and scheduling method as described in any one of claims 1 to 12.

26. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the computing power allocation and scheduling method according to any one of claims 1 to 12.

Images