Research On Key Technology Of Multi-Party Demand Scheduling

In the industrial Internet,a large number of edge nodes are deployed,and the computing resources of each edge node are different.The architecture is simple and can only handle some simple tasks.At the same time,there are a large amount of data that needs to be processed in time,and some data processing is very complicated.If you use Normal server-side processing takes a lot of time and cost.By deploying artificial intelligence models on edge nodes,complex data processing problems can be solved.However,it is often very difficult to find the optimal solution for task scheduling using previous edge computing scheduling algorithms.To solve the problem of insufficient computing power when AI models are deployed on industrial Internet terminal equipment,we propose to deploy AI models on edge server nodes as the resource supply side,and reduce the redundancy of AI models through a new mode of joint rental of node computing power and AI models and improve edge resource utilization.The main research content of this thesis is as follows:(1)In order to meet the data processing needs of end users,we have built a buyer-decisionmaking platform-seller tripartite system,and proposed a data-based,model-based,and computing power-based trade-off pricing mechanism,which is highly interpretable and uses task scheduling Reasonable,high-efficiency,and high-quality completion are the premise to achieve the goals of maximizing the overall benefits of nodes and optimizing load balancing weights.The difference from the general auction mechanism is that our system adopts the reverse auction mechanism and allows one node to compete for multiple tasks at the same time,which enables parallel processing of multiple tasks,reduces delay waiting time,and is more real-time.(2)In order to achieve the goal optimization of maximizing the overall benefits of nodes and optimizing load balancing weights,this thesis proposes a new scheduling method based on the task-decision platform-node tripartite scheduling platform.This method takes all tasks to be completed on time and with high quality as the benchmark,comprehensively schedules important indicators such as tasks,edge nodes,and global revenue,and achieves a relative balance between the interests of both tasks and nodes.The genetic algorithm is used to solve the problem,and a large number of experiments have been carried out.In a network experiment with 20 nodes and 40 tasks competing,36 tasks were completed,the task completion quality evaluation value was 0.94,the task penalty evaluation value was 0.08,the worker node historical reputation evaluation value was 0.71,and the global revenue evaluation value was0.96,which is the best compared to other methods.The experimental comparison proves that the proposed method can ensure the maximization of benefits while ensuring the high-quality and high-efficiency completion of tasks,and the load balance of the working nodes is optimal.