Research And Implementation Of Multi-objective Task Scheduling In Dispersed Network Environment

Nowadays,the Internet industry is developing rapidly and various smart devices are becoming more and more abundant.The emergence of these devices has enriched the computing resource environment,but due to their heterogeneity,mobility and dispersion,the idle resources of these devices cannot be fully utilized.Based on this,the concept of the Dispersed Network has emerged,which emphasizes the flexible collaboration among local devices with limited resources,computational and communication capabilities to accomplish complex tasks,thus making full use of the idle computational resources of these devices.Among them,how to achieve the effective management of limited resources is the main problem solved by dispersed computing task scheduling.Therefore,this paper analyzes the optimization objectives of task scheduling in a specific application scenario of dispersed network and researches the multi-objective task scheduling techniques in the dispersed network environment,the main work is as follows:Firstly,on the one hand,this paper designed and implemented a dispersed network prototype system scheduling module based on the thought of Dispersed computing,which provides the basic resource allocation capability and task scheduling capability in a dispersed network environment.On the other hand,based on the heterogeneous and highly dynamic characteristics of the Dispersed network,we construct the dispersed network environment model and the dispersed task scheduling model,which are used to describe the heterogeneous resource capability of the dispersed network environment and the various resource requirements of the tasks to be scheduled,respectively.For a specific disaster relief scenario,the dispersed network task scheduling goal is to minimize completion time and energy consumption,so the DSPSO algorithm is proposed by combining the task scheduling with a heuristic algorithm and taking into account the characteristics of dispersed network environment.The DSPSO algorithm is an enhancement of the classical PSO algorithm,which solves the multi-objective task scheduling problem in specific scenarios by converting the position of particle iterations into a scheduling policy on the one hand,and combining the task scheduling optimization objective with the fitness function on the other hand.Finally,the iterative optimization effect of the algorithm and the optimization effect on the scheduling objectives of task completion time and energy consumption as well as the practical application value of the algorithm are verified through simulation experiments and system applications.For a specific scenario of collaborative processing of military image intelligence under the battlefield,the task scheduling optimization objectives for this scenario are to minimize task completion time and maximize resource utilization.Therefore,the DSDRL algorithm is proposed by combining the task scheduling problem with reinforcement learning and making full use of the interaction of the reinforcement learning Agent with the abstract environment.The DSDRL algorithm abstracts the task scheduling problem in a dispersed network environment into a reinforcement learning environment on the one hand,and transforms the task scheduling optimization objective into a reward mechanism on the other hand,thus realizing its self-learning and self-adaptation.Finally,the training effect of the algorithm and the optimization effect for the task scheduling objective as well as the practical application value of the algorithm are verified through simulation experiments and system applications.