Search

CN-122027583-A - Contribution-degree-based edge computing node scheduling method and system

CN122027583ACN 122027583 ACN122027583 ACN 122027583ACN-122027583-A

Abstract

The application discloses a contribution-based edge computing node scheduling method and system, which relate to the technical field of edge computing and comprise the steps of constructing a reference contribution potential model based on historical contribution by collecting operation state information such as computing frequency, bandwidth, energy consumption and the like of nodes and combining with historical task performance of the nodes to realize quantitative evaluation of long-term capability of the nodes; the method comprises the steps of constructing a topological graph by utilizing the adjacent relation among nodes, performing diffusion operation on reference potential, spreading the cooperative influence of high-contribution nodes to the neighborhood nodes, constructing a comprehensive scheduling energy function comprising time delay, energy consumption, diffusion potential and fairness punishment items by combining task requirements and node characteristics in a task allocation stage and predicting operation time delay and energy consumption, and forming a self-adaptive feedback closed-loop mechanism by dynamically updating the historical contribution of the nodes according to the actual execution result of the tasks. Therefore, low-time-delay, high-energy-efficiency and long-term fair task scheduling can be realized in a multi-node heterogeneous environment.

Inventors

  • GAO HUI

Assignees

  • 鹏博士大数据有限公司

Dates

Publication Date
20260512
Application Date
20260204

Claims (9)

  1. 1. An edge computing power node scheduling method based on contribution degree, which is characterized by comprising the following steps: collecting node running state information and historical contribution degree of each edge node, and calculating reference contribution potential of the corresponding edge node according to the historical contribution degree, wherein the node running state information comprises node calculation frequency, node available bandwidth and unit calculation energy consumption coefficient; Constructing a topological graph according to the adjacent relation of each edge node, and performing diffusion operation on the reference contribution potential to generate diffusion contribution potential of each edge node; Predicting expected operation time delay and expected energy consumption of a target computing task on each edge node based on task requirements of the target computing task to be distributed and node operation state information of each edge node; Calculating comprehensive dispatching energy function values according to the expected operation time delay, the expected energy consumption, the diffusion contribution potential and fairness penalty items corresponding to all edge nodes, and selecting the edge node with the smallest comprehensive dispatching energy function value as a task bearing node; and issuing the calculation task to the task bearing node, and after the calculation task is completed, updating the historical contribution degree of the task bearing node according to the actual running time delay and the actual running energy consumption corresponding to the calculation task.
  2. 2. The method of claim 1, wherein updating the historical contribution of the task-carrying node based on the actual operational delay and the actual operational energy consumption corresponding to the computing task comprises: Acquiring actual running time delay of completing calculation task of task bearing node g at time m And actual energy consumption To calculate task performance indicators corresponding to the computing tasks: , In the formula, For the time delay energy consumption weighting factor, Task performance indexes for finishing the calculation task at the moment m for the edge node g; Updating the historical contribution degree of the task bearing node according to the task performance index: , In the formula, And Represents the historical contribution of the task carrying node g at time m and time m-1 respectively, Is the attenuation coefficient.
  3. 3. The method of claim 1, wherein said calculating reference contribution potentials for respective edge nodes from said historical contribution comprises: Normalizing according to the historical contribution degree of each edge node in the edge network to obtain the reference contribution potential of the corresponding edge node: , where N is the total number of edge nodes in the edge network, And To represent the historical contribution of the ith and the ith edge node at time t respectively, Representing the reference contribution potential of the edge node i at time t.
  4. 4. The method of claim 1, wherein constructing a topology graph from the proximity of each edge node and performing a diffusion operation on the reference contribution potential to generate a diffusion contribution potential for each edge node comprises: integrating reference contribution potentials of respective edge nodes in an edge network to construct a reference contribution potential vector : , In the formula, Representing the reference contribution potential of the nth edge node at time t, Representing a vector transpose operation; constructing an undirected topology graph Assigning edge weight to each edge And normalizing the adjacent matrix according to the rows to obtain a diffusion weight matrix ; As a set of edge nodes, Is an edge set; Representing edge weights between an ith edge node and a jth edge node, which are calculated according to the cooperative relationship among the nodes, the computing capability difference, the link quality and the historical task interaction frequency; initialization with reference contribution potential vector And adopts a diffusion iteration model: , Wherein, the For the diffusion coefficient, q represents the number of iterations, Representing a diffusion contribution potential vector after the q-th iteration; When the iteration number exceeds a preset number threshold, a steady-state diffusion contribution potential vector is obtained : , In the formula, And representing steady-state diffusion contribution potential values of the Nth edge node, wherein the steady-state diffusion contribution potential values are used for representing global cooperation contribution degree of the node in the whole topological network.
  5. 5. The method of claim 4, wherein each edge is given an edge weight And normalizing the adjacent matrix according to the rows to obtain a diffusion weight matrix Comprising: , In the formula, Representing the physical distance between the i-th edge node and the j-th edge node, In the form of an edge set, Representing the available bandwidths of the i-th and j-th edge nodes respectively, Indicating the reliability of the link communication between the ith and jth edge nodes, And The historical task cooperation success times and failure times of the ith edge node and the jth edge node are respectively, As a parameter of the distance scale, In order to set the constant value of the preset value, Is a weighted index coefficient; Normalizing the edge weight matrix according to the rows to obtain a diffusion weight matrix : , In the formula, Representing a diffusion weight matrix The element of the ith row and the jth column is used for describing a normalized weight coefficient of the contribution potential of the ith edge node to diffuse to the jth edge node; Representing the total number of neighboring edge nodes to the ith edge node, a represents the node index of the neighboring edge node, Representing the sum of the edge weights between the i-th edge node and all neighboring edge nodes.
  6. 6. The method according to claim 1, wherein predicting the expected running delay and the expected energy consumption of the target computing power task on each edge node based on the task requirement of the target computing power task to be allocated and the node running state information of each edge node comprises: Extracting a target calculation task Task demand parameters of (a) Wherein Representing the calculated amount of the task, Representing the amount of task input data and extracting the operation state parameters of the ith edge node Wherein The node calculation frequency representing the i-th edge node, The node available bandwidth representing the ith edge node, A unit calculation energy consumption coefficient representing an ith edge node; based on heterogeneous resource characteristics and task characteristics of the nodes, calculating predicted running time delay and predicted energy consumption: , , In the formula, Representing a target calculation task The predicted run time delay at the ith edge node, Representing a target calculation task Predicted energy consumption at the ith edge node, The method is used for adjusting the influence of communication bandwidth on time delay for the transmission delay weight factor; for calculating the energy consumption conversion coefficient, the method is used for quantifying the influence of the calculated amount on the energy consumption; is the equivalent average active power of the ith edge node during task execution.
  7. 7. The method of claim 1, wherein the calculation of the fairness penalty term comprises: Setting the task allocation quantity vector of the edge network as And calculates the average task number of the system : , In the formula, And Respectively representing the task allocation numbers of the ith edge node and the Nth edge node; generating corresponding fairness penalty items for each edge node respectively: , Wherein, the A fairness penalty value representing an ith edge node, To be a fair penalty coefficient, denominator Preventing zero value from overflowing when Time of day Indicating that the task load of the ith edge node is higher than the average level of the system, the scheduling algorithm will automatically increase the penalty value of the node, thereby inhibiting the scheduling algorithm from continuing to allocate new tasks to the node when Time of day The task load of the ith edge node is lower than the average level of the system, and the scheduling algorithm automatically reduces the scheduling cost of the ith edge node, so that the scheduling algorithm is encouraged to migrate part of tasks to the ith edge node, and the balance of overall task allocation is improved.
  8. 8. The method of claim 7, wherein said calculating an integrated scheduling energy function value based on the expected operating delay, the expected energy consumption, the diffusion contribution potential, and a fairness penalty term for each edge node comprises: , In the formula, Representing a target calculation task The predicted run time delay at the ith edge node, Representing a target calculation task Predicted energy consumption at the ith edge node, Representing the steady state diffusion contribution potential of the ith edge node, A fairness penalty value representing an ith edge node, Is a normalized weight coefficient; Representing a target calculation task An integrated dispatch energy function value at an ith edge node, wherein The lower the i-th edge node is performing the target computing power task The smaller the composite cost.
  9. 9. An edge computing power node scheduling system based on contribution, the system comprising: The contribution potential initializing unit is used for collecting node running state information and historical contribution degrees of all edge nodes and calculating reference contribution potential of the corresponding edge nodes according to the historical contribution degrees, wherein the node running state information comprises node calculation frequency, node available bandwidth and unit calculation energy consumption coefficient; The contribution potential diffusion unit is used for constructing a topological graph according to the adjacent relation of each edge node, and performing diffusion operation on the reference contribution potential to generate diffusion contribution potential of each edge node; the task performance prediction unit is used for predicting expected operation time delay and expected energy consumption of the target computing task on each edge node based on task requirements of the target computing task to be distributed and node operation state information of each edge node; The task scheduling analysis unit is used for calculating comprehensive scheduling energy function values according to the expected operation time delay, the expected energy consumption, the diffusion contribution potential and fairness penalty items corresponding to the edge nodes, and selecting the edge node with the minimum comprehensive scheduling energy function value as a task bearing node; and the task issuing unit is used for issuing the force calculation task to the task bearing node, and updating the historical contribution degree of the task bearing node according to the actual running time delay and the actual running energy consumption corresponding to the force calculation task after the force calculation task is completed.

Description

Contribution-degree-based edge computing node scheduling method and system Technical Field The application relates to the technical field of edge computing, in particular to an edge computing power node scheduling method and system based on contribution degree. Background In recent years, digitization and networking have been in progress, so that a large number of sensing devices continue to generate massive amounts of data. Traditional cloud computing centers, while having powerful computing capabilities, have difficulty meeting the requirements of low latency applications due to the large physical distances, network congestion, and central load fluctuations. In view of this, an edge computing architecture that deploys computing resources at the edge of the network has emerged, which significantly reduces the backhaul latency and bandwidth consumption of perceived data by processing the data nearby. Edge nodes include both operator deployed micro data centers and computing power resources provided by the interaction between end users. Due to the large number of nodes, strong isomerism and limited available computing power for most nodes, how to schedule these nodes reasonably becomes a key issue. Traditional scheduling schemes are often designed around reducing average delay and energy consumption, which emphasize real-time performance, but neglect balance of paying out and earning of each node in long-term operation, so that the phenomenon that resources are occupied by a few high-performance nodes for a long time exists. Disclosure of Invention The application provides a contribution-based edge computing power node scheduling method, a contribution-based edge computing power node scheduling system, a storage medium, a computer program product and electronic equipment, which are used for at least solving the structural imbalance problem that the low-delay service requirement and the long-term fair utilization of multi-node resources are difficult to be compatible in the edge computing scene in the prior art. According to a first aspect, the embodiment of the application provides an edge computing power node scheduling method based on contribution degree, which comprises the steps of collecting node operation state information and historical contribution degree of each edge node, calculating reference contribution potential of the corresponding edge node according to the historical contribution degree, wherein the node operation state information comprises node calculation frequency, node available bandwidth and unit calculation energy consumption coefficient, constructing a topological graph according to the adjacent relation of each edge node, performing diffusion operation on the reference contribution potential to generate diffusion contribution potential of each edge node, predicting expected operation time delay and expected energy consumption of a target computing power task on each edge node based on task demand of a target computing power task to be distributed and node operation state information of each edge node, calculating comprehensive scheduling energy according to the expected operation time delay, expected energy consumption, the diffusion contribution potential and fairness penalty function value corresponding to each edge node, selecting the edge node with minimum comprehensive scheduling energy as a task bearing node, wherein the calculation cost is calculated according to the deviation of the distributed task number of the edge node and the average task distribution number of all nodes of a system, and is used for inhibiting the actual contribution power of the task from being distributed to the actual computing power of the corresponding computing task, and the actual contribution power is completed after the actual contribution power is balanced. In a second aspect, the embodiment of the application provides an edge computing power node scheduling system based on contribution, which comprises a contribution potential initializing unit, a task performance predicting unit, a task scheduling analyzing unit and a task scheduling and analyzing unit, wherein the contribution potential initializing unit is used for collecting node operation state information and historical contribution degrees of all edge nodes and calculating reference contribution potentials of corresponding edge nodes according to the historical contribution degrees, the node operation state information comprises node calculation frequency, node available bandwidth and unit calculation energy consumption coefficients, the contribution potential diffusing unit is used for constructing a topological graph according to the adjacent relation of all edge nodes and performing diffusion operation on the reference contribution potentials to generate diffusion contribution potentials of all edge nodes, the task performance predicting unit is used for predicting expected operation time delay and expected energy consumption of the