CN-121996388-A - Cloud edge cooperative task scheduling method based on optimization A2C algorithm

CN121996388ACN 121996388 ACN121996388 ACN 121996388ACN-121996388-A

Abstract

The invention discloses a cloud edge cooperative task scheduling method based on an optimization A2C algorithm, and belongs to the technical field of cloud computing and edge computing fusion scheduling. Aiming at the technical problems of low resource utilization rate and high system energy consumption and delay of the existing cloud edge collaborative task scheduling method, the invention provides a method for constructing a task-host heterogram, explicitly modeling bidirectional constraint and competition relation between task demands and host resources, designing a heterogeneous graph pointer network, respectively extracting competition strength of tasks to the host and supply and demand matching quality of the tasks and the host through a double-stage graph attention mechanism to form a joint environment state characteristic, adopting a task granularity dominance function, differentially calculating dominance values of each task action according to scheduling execution results and global time sequence differencing errors, and carrying out strategy network gradient updating by combining pointer probability. The method realizes accurate perception and fine granularity optimization of the dynamic cloud edge environment, and remarkably reduces the energy consumption of the system and the task response delay.

Inventors

ZHANG XIUGUO
AN XINRU
WANG XI
CAO ZHIYING

Assignees

大连海事大学

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (8)

1. A cloud edge cooperative task scheduling method based on an optimization A2C algorithm is characterized by comprising the following steps: Receiving a task set to be scheduled in a discrete time interval, and constructing a cloud edge cooperative task scheduling model by taking the minimum overall energy consumption of a cloud edge host in one time interval and the average response time of tasks completed to be executed in the time interval as optimization targets; Constructing a task-host heterogram representing a mapping relation between task resource requirements and host resource supply and resource constraint conditions, wherein the task-host heterogram comprises a task node set, a host node set and a bidirectional edge set with direction semantics, wherein a forward edge represents a task which can be distributed to a host node, and a reverse edge represents a host node which can bear the task; Solving a cloud edge collaborative task scheduling model based on an optimized A2C network, wherein the optimized A2C network comprises a state, a task-host heterogram, a heterogeneous graph pointer network, an environment, a value network and a loss layer, and the state part acquires a real-time environment state at the starting moment of a current scheduling interval; the task-host heterograph part constructs a task-host heterograph according to the environmental state; the heterogeneous graph pointer network part captures the competition relationship and the matching relationship of tasks to host resources through a graph convolution layer, and further dynamically generates task scheduling decisions at a decision layer; the environment part provides environment feedback information and a new environment state after executing the decision, acquires the environment feedback information and the environment state at the next moment after executing the task scheduling decision, calculates the loss of the strategy network and the value network based on the optimized dominance function and is used for updating the learnable parameters of the optimized A2C network; When both the loss of the value network and the loss of the policy network converge, an optimal task scheduling decision is obtained that minimizes the system energy consumption and delay.
2. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 1, wherein the task-host heterogram construction comprises: For any task node and host node, a bi-directional edge is established between them if and only if the following four resource constraints are satisfied at the same time: the task computing requirement is smaller than the available computing resources of the host node; the memory requirement of the task node is smaller than the available memory resource of the host node; the bandwidth requirement of the task node is smaller than the available bandwidth resource of the host node; the current load rate of the node is lower than a preset threshold.
3. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 1 is characterized in that the heterogeneous graph pointer network part captures the competition relation and the matching relation of tasks to host resources through a graph convolution layer, and the cloud edge collaborative task scheduling method comprises the steps of introducing an attention mechanism into a graph convolution layer of the heterogeneous graph pointer network, extracting environment state characteristics of task-host heterograms input into the heterogeneous graph pointer network, firstly extracting competition relation characteristics of tasks to host resources based on task-host edge joint node information, and extracting matching relation characteristics of task requirements and host resources based on host-task edge joint node information and competition relation characteristics.
4. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 3, wherein the heterogeneous graph pointer network part captures the competition relationship and the matching relationship of tasks to host resources through a graph convolution layer, and the cloud edge collaborative task scheduling method further comprises the step of splicing task node characteristics and matching relationship characteristics to obtain environment state characteristics of an affiliated time interval.
5. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 4, wherein the heterogeneous graph pointer network dynamically generates task scheduling decisions at a decision layer, and the cloud edge collaborative task scheduling method comprises the steps of processing extracted environmental state features by using a multi-layer perceptron to generate probability distribution of task scheduling to each host; By using And the strategy dynamically generates task scheduling decisions according to the task scheduling probability distribution.
6. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 5, wherein the optimized dominance function is Calculated as follows: When (when) And is also provided with In the time-course of which the first and second contact surfaces, ; When (when) And is also provided with In the time-course of which the first and second contact surfaces, ; When (when) In the time-course of which the first and second contact surfaces, ; Wherein the method comprises the steps of As a result of the time-series differential error, For time intervals of Is used for the optimization of the target value of (c), For a subset of task scheduling decisions that have not been successfully performed due to resource contention or insufficient resources, To take the value in the range of Scaling factors within.
7. The cloud edge collaborative task scheduling method based on the optimization A2C algorithm according to claim 6, wherein the loss function of the policy network is: Wherein, the For the task Is recorded with a task The probability of being scheduled to the target host, Taking logarithm for probability; the loss function of the value network is: Wherein, the Is a time series differential error.
8. The cloud edge collaborative task scheduling method based on the optimized A2C algorithm according to claim 7, wherein updating the optimized A2C network learnable parameters comprises: And obtaining respective gradients by deviant guiding all parameters through counter-propagation according to the losses of the value network and the strategy network, and updating the parameters through gradient descent.

Description

Cloud edge cooperative task scheduling method based on optimization A2C algorithm Technical Field The invention relates to the technical field of computers, in particular to a cloud edge cooperative task scheduling method based on an optimization A2C algorithm. Background Currently, technologies such as 5G, internet of things, cloud computing, edge computing and the like are rapidly developing. The application program is run on the terminal of the Internet of things, and data acquired by the terminal equipment such as the sensor are processed and analyzed in real time through the cloud edge cooperative technology, so that intelligent decision support can be provided for a terminal user, and the intelligent Internet of things can be further realized. In the cloud edge cooperative environment, as the types, the number and the resource requirements of the computing tasks and the network bandwidth between the data sources and the computing nodes are continuously and dynamically changed, and meanwhile, the edge hosts and the cloud processing center also have the characteristics of geographic dispersion, resource and equipment isomerism and the like, the response time and the energy consumption of different hosts when processing different types of computing tasks are obviously different, and how to dynamically allocate the computing tasks generated by the Internet of things terminal to the optimal edge nodes or cloud nodes by utilizing an intelligent scheduling strategy, so that efficient computing task scheduling is realized, and the cloud edge cooperative environment becomes a research hotspot. The task scheduling decision is a NP difficult problem under cloud-edge cooperative environment with complex structure and dynamic change, and the optimal solution can not be obtained in polynomial time, so that the existing work is mostly solved by using a mathematical programming method, a heuristic algorithm, a deep reinforcement learning algorithm and the like to establish a task scheduling model. (1) Cloud edge cooperative task scheduling method based on mathematical programming The method is based on a mathematical principle, and can accurately express an objective function and constraint conditions of task scheduling by constructing a mathematical model, so that a global optimal solution is obtained theoretically. For example, in a cloud-edge collaborative environment, the task scheduling and resource allocation problems are formulated as Mixed-Integer nonlinear Programming (MINLP), the MINLP is converted into Mixed-Integer linear Programming (millp) by piecewise linear approximation and linear relaxation, a near-optimal solution is obtained through a gap-adjusted branch-and-bound algorithm, and efficient scheduling is realized in a large-scale task scenario. However, due to the dynamic change of cloud-edge collaborative environment, a large number of assumptions and constraint conditions are often needed in such research, the model construction is complex and ideal, the model solving is relatively difficult, and the dynamic fluctuation of the scene is difficult to cope with. (2) Cloud edge cooperative task scheduling method based on heuristic algorithm The method utilizes a predefined and empirical intuitive rule set to make scheduling decisions or simulate natural and physical processes, and gradually approaches better solutions through iterative improvement. For example, heuristic rules are used for rapidly judging task demands and host states of cloud edge collaborative environments, task scheduling decisions are made according to the rules, genetic algorithms encode the task scheduling decisions into chromosomes, population individuals are evolved through selection, intersection, mutation and other operations, an approximately optimal solution is obtained, ant colony algorithms simulate task scheduling processes into ant selection paths, optimal paths are found through pheromone positive feedback, simulated annealing algorithms set high initial temperatures for the task scheduling decisions, multiple candidate solutions are effectively explored in a huge solution space through controlling cooling rates, and finally global optimization tends to be achieved. However, since the nature of heuristic algorithms is a trade-off between speed and accuracy, the solution provided by heuristic algorithms in fast decision-making is often a locally optimal or near optimal solution for single-objective optimization, usually considering only the current state and the decisions between multiple tasks are relatively isolated, lacking global view and long-term optimization. If the local optimization is skipped and the global optimization is considered, the decision time and the calculation cost of the heuristic algorithm are greatly increased, so that the heuristic algorithm has a further optimization space in a cloud edge cooperative task scheduling scene. (3) Cloud edge cooperative task scheduling method based on deep reinfor