Research And Implementation Of Task Management And Resource Monitoring Technology In Dispersed Computing

Dispersed computing is a new computing paradigm proposed in recent years,which aims to build a decentralized and self-organized mobile communication network using resourceconstrained nodes,so that a number of mobile nodes form a computing organism and coordinate computing tasks using node resources.The dispersed ad hoc network studied in this paper is a network model in the context of dispersed computing,and this paper focuses on the resource monitoring technology and task management technology based on resource monitoring under this network model.Based on the research of existing resource monitoring and task management technologies,this paper designs and implements a resource monitoring subsystem and a task management subsystem in the dispersed ad hoc network environment,which solve the problems of difficult resource state sensing and complex task collaborative computation under the dispersed ad hoc network,respectively.The paper will focus on the design and implementation of resource monitoring and task management techniques.In terms of resource monitoring,firstly,for the decentralized and self-organized characteristics of dispersed ad hoc networks,this paper proposes a clustered resource monitoring logic structure.Secondly,combined with the shortcomings of the existing clustering algorithms,a weighted clustering algorithm suitable for decentralized selforganizing networks is proposed.Through simulation tests,it is verified that the proposed clustering algorithm can improvc the stability and scalability of the network.Based on the proposed weighted clustering algorithm,cluster head election and node entry are completed to build a stable and efficient clustering structure,while a backup cluster head mechanism,a cluster head replacement mechanism and a node dynamic maintenance mechanism are designed and implemented to maintain the relative stability of the clustering structure.Then,the resource data collection module is designed and implemented to complete the resource data collection of nodes.Finally,the intra-cluster data synchronization mechanism with incremental updates and the inter-cluster data synchronization mechanism based on the improved Gossip algorithm are designed and implemented to synchronize the resource data within and between clusters.By analyzing the redundancy problems and causes of redundancy in the traditional Gossip algorithm,the traditional Gossip algorithm is improved,and the improved Gossip algorithm is verified to improve data synchronization efficiency and reduce data synchronization redundancy by analyzing message redundancy and data synchronization period and other indicators through simulation tests.After testing the resource monitoring subsystem,the results show that the dispersed ad hoc network can effectively build a stable sub-cluster monitoring structure,and the resource collection implementation and data synchronization implementation can guarantee the effective sensing and sharing of resource data among computing nodes.In terms of task management,firstly,a decentralized task management architecture is designed for the decentralized and self-organized characteristics of dispersed ad hoc networks.Secondly,we design a unified dynamic task management scheme and propose a task registration mechanism based on Java annotation technology,based on which we design a task visualization management platform to realize the unified dynamic management of different tasks.Then,we design the task scheduling,task execution and task monitoring scheme,decouple the task "scheduling process" and task "execution process",and design and implement the resolution,scheduling,execution and monitoring of DAG workflow tasks based on the resource data provided by the resource monitoring system.Finally,this paper designs a high availability scheme for task scheduling and task execution.On the one hand,the scheduling rights of tasks with scheduler failure are preempted by the election mechanism based on Raft protocol to guarantee the high availability of task scheduling.;on the other hand,the task failure retry mechanism is designed to solve the problems of task runtimeout,executor failure and final task execution failure.After testing the task management subsystem,the results show that the system meets the task management requirements well and achieves the design objectives.