| With the rapid development of the Internet of Things(Io T),an increasing number of intelligent terminal devices are being connected to network environments.In order to overcome the high end-to-end latency,channel transmission pressure,and user information security risks associated with centralizing all data processing and storage in cloud data centers,edge computing has emerged.As an extension of cloud computing,edge computing typically leverages edge servers located near the terminal devices to provide computing and storage capabilities.This reduces data transmission time,alleviates channel transmission pressure,lowers latency,bandwidth consumption,and ensures data security more effectively.However,compared to cloud servers,edge servers still have limited resources and suffer from performance limitations,making them unsuitable for complex tasks such as deep learning.Therefore,it is necessary to integrate cloud computing and edge computing,and one major challenge in this integration is how to effectively schedule and allocate resources in the cloud-edge environment.This thesis focuses on studying the resource coordination problem in the cloud-edge environment,and the main contributions are summarized as follows:(1)Addressing the network congestion issue in traditional central cloud monitoring systems,this thesis proposes a resource monitoring architecture for the cloud-edge environment.A new component called Edge Mon is developed based on Node Exporter and Pushgateway.Edge Mon can temporarily store metric data at the edge layer and evaluate these metrics using a rule engine.Once triggered by the rule engine,Edge Mon sends alert messages to the cloud.Evaluation results in a comparative scenario demonstrate that the monitoring system composed of Edge Mon can provide real-time metrics and alerts with minimal edge requirements,effectively avoiding network congestion.(2)Addressing the resource scheduling problem in the cloud-edge collaborative computing environment,this thesis formulates the task scheduling and resource allocation problem as a multi-objective optimization problem aiming to minimize execution time,scheduling cost,and average load while respecting deadline constraints.An improved algorithm based on multi-objective particle swarm optimization is adopted to solve the problem.The algorithm combines importance strategy and multi-objective particle swarm optimization to select the global best leader and introduces a mutation operator to prevent premature convergence to local optima.Comparative experiments are conducted,comparing the improved algorithm with the original algorithm and similar algorithms.The results show that the proposed algorithm outperforms traditional multi-objective particle swarm optimization and multi-objective genetic algorithm,effectively reducing scheduling costs,decreasing task execution time,and ensuring cluster load balancing.(3)Addressing the task offloading and resource allocation problem in the cloud-edge environment,this thesis abstracts the collaborative offloading problem as a Markov decision process.Partial offloading strategies are devised for dependent tasks based on a directed acyclic graph task model.A mathematical model is established to minimize the overall system cost(a weighted combination of latency and energy consumption).Considering the complexity and dynamics of the cloud-edge environment,this thesis adopts an improved algorithm called DQNRC,based on deep reinforcement learning using deep Q-networks,to solve the optimal offloading vector.Extensive simulation experiments are conducted to validate the effectiveness of the proposed algorithm. |
