DNN Application Computing Migration Decision Technology Based On Heuristic Algorithm In Edge Environment

Deep Neural Networks(DNN)are widely used in many applications.The traditional method is to migrate them to the cloud,which will cause serious data transmission delays,high cost of network resources,and user privacy leaks.The emergence of mobile edge computing provides a new solution for the execution of DNN-based applications.Migrating some layers of DNN applications to the edge can speed up system response and reduce the burden on cloud centers.However,it is difficult to make computing migration decision for DNN application due to the complexity of DNN application and edge environment.There are two challenges for DNN applications in computing migration decisions in edge environments.On the one hand,it comes from cost optimization.Unlike the traditional cloud computing,the pricing model of edge servers is more complex and changeable.On the other hand,it comes from energy consumption optimization.Different DNN application computing migration schemes will lead to different energy consumption.An unreasonable migration scheme will greatly increase energy consumption.In response to these challenges,the main work is as follows:(1)Establish an edge network topology environment model including Io T device end,edge node end and cloud,make constraints on the problem according to real scenarios and construct a unified model of DNN application migration in the edge environment according to different optimization goals.From the model level the problem to be solved in this article is explained.(2)On the problem of cost-optimized DNN application migration,various influencing factors such as the application structure of different DNNs and the characteristics of different types of computing nodes on the total cost of the DNN application migration system is considered.In response to this problem,a DNN application migration decision-making technology for cost optimization is proposed.By introducing the crossover operator and mutation operator of genetic algorithm into the discrete particle swarm algorithm,this technology effectively avoids the problem of premature convergence of the particle swarm algorithm,and can effectively reduce the total system cost of DNN application migration.(3)On the problem of DNN application migration for energy consumption optimization,considering the influence of computing node operating interval and other factors on the total energy consumption of computing nodes,a DNN application migration decision-making technology for energy consumption optimization is proposed for this problem.This technology is based on the discrete particle swarm algorithm with the introduction of genetic operators and adjusts the algorithm parameters adaptively to further avoid the premature convergence problem,and can effectively reduce the total energy consumption of the DNN application migration system.Finally,this paper simulates the relevant experimental environment and evaluates the proposed technology to verify the effectiveness of the technology.The result show that the cost-optimized DNN application computing migration decision-making technology can reduce the cost by 19% to 27% compared with other technologies,and the energy-consuming optimization-oriented DNN application computing migration decision-making technology can reduce the energy consumption 13% to 24% compared with other technologies.