Research On Collaborative Offloading Strategy In Cloud-edge Collaborative Network

Benefiting from the powerful network capabilities brought by the Fifth Generation Mobile Communication Technology and the Sixth Generation Mobile Communication Technology,traditional industries are rapidly transforming to digitalization driven by Artificial Intelligence.Intelligent services have diverse computing requirements,and it also makes it difficult for existing Multi-access Edge Computing and Mobile Cloud Computing to meet the differentiated requirements of intelligent applications for indicators such as low latency and low energy consumption.As one of the key technologies of mobile communication networks,cloud-edge collaboration,which combines the rich resource supply of cloud servers and the flexible service capabilities of edge servers,is an emerging technical solution to address the above computing requirements.Compared with MEC and MCC,cloud-edge collaboration has a more complex network structure,more flexible resource allocation,and richer options for offloading strategies.Therefore,the optimization of collaborative offloading strategies is the core issue.By formulating an effective collaborative offloading strategy,it is possible to make full use of the resources of cloud servers and edge servers,release network capabilities,meet the differentiated needs of users,allocate resources in a timely and effective manner,reduce computing costs of users,and improve service experience of users.To sum up,this thesis mainly studies the offloading scheme in the cloud-edge collaborative network,and proposes two collaborative offloading strategies and algorithms for system overhead optimization and other issues.First,we propose a parallel deep learning-driven collaborative offloading scheme for cloud-edge collaborative network.Firstly,we propose two edge computing modes based on cloud-edge collaboration,and build models for the latency and energy consumption required by user computing tasks under different computing modes.Secondly,we propose an offloading algorithm based on parallel deep learning with cost of users as the optimization goal,which considers the processing problem of unsupervised learning and the convergence speed of the algorithm.Finally,the simulation results show that the proposed strategy can generate suboptimal offloading decisions in a relatively short time,which effectively reduces the computing cost of users.Second,we propose a pricing-driven positive-sum game cooperative offloading strategy.Firstly,we design a back-off mechanism and pricing scheme based on cloud-edge collaboration,in which the resource utilization of the server is fully considered.Secondly,we build models for users’ latency,energy consumption,and payment costs.Taking the user cost as the optimization goal,we reconstruct the cooperative offloading problem into a positive-sum game problem,and prove the existence of Nash Equilibrium.Thirdly,we propose an offloading algorithm for positive-sum games based on dynamic pricing.Finally,the simulation results show that the proposed offloading strategy can enable more users to participate in offloading,effectively alleviate the problem of insufficient computing resources in the network,and reduce the average cost of users.