| In recent years,with the rapid development of 5G communication technology and Io T,mobile data has shown an explosive growth trend.Compared with the traditional cloud-centered architecture,the cloud-edge-terminal collaborative network architecture combines the two major advantages of low processing delay of mobile edge computing and powerful computing capability of mobile cloud computing,which reflects higher timeliness and reliability in processing mobile data generated by latency-sensitive and computationally intensive applications.As two key technologies in the cloud-edge-terminal architecture,computation migration and computation offloading play an important role in ensuring server load balancing,reducing computation task execution delay,lowering system energy consumption,and improving the user service quality.However,most computation migration methods currently overlook the instability of mobile edge networks when executing migration work.In reality,network failure factors have varying degrees of impact on migration work between edge servers.As for computation offloading,most current works focus on the overall offloading of computation tasks,neglecting the dependence relationship between computation tasks.Moreover,partial offloading methods based on optimization algorithms generally encounter the problem of high complexity in handling multi-terminal device offloading.To address the above issues,this thesis studies the computation migration and offloading problems in the cloud-edge-terminal collaborative scenario,and the main work is as follows:In terms of computation migration,this thesis proposes a computation migration method based on mobile edge network failure prediction.The method introduces SDN as a global and local controller for computation migration at the mobile edge layer and central cloud layer,respectively,to monitor and summarize the edge network situation and perform migration task scheduling.Secondly,this thesis proposes a network failure prediction model based on an improved Wide&Deep model to predict imminent network failures based on network device alarm information.Finally,the computation migration problem is constructed as a Markov decision process,and a failure penalty function is designed to avoid failure nodes.The optimal migration strategy is solved using the deep Q-learning method in deep reinforcement learning.Experimental results verify the effectiveness of the proposed network failure prediction method and computation migration method.In terms of computation offloading,this thesis proposes a partial computation offloading method based on task dependency relationships.The method first constructs a cloud-edge-terminal collaborative computation offloading system architecture and establishes a dependency relationship and priority model for computation tasks.Then,the model unloads subtasks to appropriate nodes based on the dependency relationship of each computation task,and establishes a partial offloading model for computation offloading with the comprehensive optimization objective of minimizing system delay and energy consumption under the three-layer architecture of cloud-edge-terminal.Finally,the soft Actor-Critic algorithm in deep reinforcement learning is used to solve the optimal offloading problem in a multi-terminal offloading environment.Simulation experiments show that the proposed computation offloading method is superior to the baseline method in reducing delay and cost and improving the application completion rate. |
