Research Of Computation Offloading Strategy Optimization Algorithms In Edge Computing

Edge Computing(EC)is the key technology in the last mile of artificial intelligence technology,and the optimization of computation offloading strategy is the core issue of EC.To improve the quality of service and the experience of user,in view of the competitions between user equipments(UEs),the mobility of UEs,the limited resources of EC,and the characteristics of End-Edge-Cloud collaborative computing(EECCC)architecture in EC,this paper proposes a series of computation offloading strategy optimization algorithms,which effectively solves the task scheduling and resource allocation problems in the heterogeneous EC environment.The main innovative research results of this paper are as follows:1.A computation offloading strategy optimization algorithm in the multiple users and multiple servers scenario.To study the computation offloading strategy optimization problem in the multiple users and multiple edge servers(ESs)scenario,this chapter first formulates the computation offloading strategy optimization problem in EC as the task overhead minimization problem under the resource and delay constraints.Subsequently,this chapter uses the variable substitution technology to transform the original non-convex optimization problem into a convex optimization problem,and proves that the problem has an optimal solution.Then,this chapter uses the theory and method of convex optimization to propose a computation offloading strategy optimization algorithm for the single UE and single ES scenario.Meanwhile,to handle the conflicts when multiple UEs initiate requests to the same ES and improve decision-making efficiency,this chapter designs a dictionary data structure to record the best offloading strategies when the UEs request each ES to perform their tasks.Then,based on the algorithm for the single UE and single ES scenario and the record dictionary,this chapter develops a distributed computation offloading strategy optimization algorithm for the multiple UEs and multiple ESs scenario.Experimental results show that the proposed algorithm has better performance than the other computation offloading strategy optimization algorithms.Meanwhile,comprehensively considering the ES resource utilization and user requirements,the experiments find that the larger the ES scale is not always the better,but should be determined according to the factors,such as user scale,application type,and network environment.2.The computation offloading strategy optimization algorithms for mobile users in the long-term and short-term.To study the computation offloading strategy optimization problem for mobile users,according to the criteria of whether the UE’s future movement location is predictable,this chapter first classifies the mobility of UEs into three types,i.e.,random mobility,predictable mobility,and known mobility.Subsequently,this chapter develops three greedy strategy based computation offloading strategy optimization algorithms for the above three types of mobility.However,it is a huge challenge to accurately predict the mobile characteristics of the UEs over a long time,especially in a highly random environment like EC.Finally,to address the issue,this chapter uses the Lyapunov optimization method(LOM)to decompose the original problem into a series of real-time optimization sub-problems,and proposes a LOM-based computation offloading strategy optimization algorithm that does not require the prior knowledge of UE mobility.Experimental results show that the greedy strategy based computation offloading strategy optimization algorithms can further reduce the UEs’ energy consumption and time consumption by using the mobility characteristics of the UEs,and the performance is better than the LOM-based computation offloading strategy optimization algorithm in the short-term.However,from a longterm perspective,the LOM-based computation offloading strategy optimization algorithm reduces the effect of instantaneous movement on the computation offloading strategy,so it is better than the greedy strategy based computation offloading strategy optimization algorithms.3.The computation offloading strategy optimization algorithms under the condition of edge service provider is cost-constrained.To improve the quality of service and the experience of user while controlling the cost of edge service provider(ESP),this chapter first formulates the computation offloading strategy optimization problem as the task completion time minimization problem under the cost constraint of ESP.Then,this chapter uses LOM to transform the original problem into a series of real-time linear programming sub-problems,develops a LOM-based computation offloading strategy optimization central algorithm,and theoretically analyzes the performance boundary of the algorithm in detail.However,these sub-problems are still NP-hard problems and the central algorithm cannot cope with the large-scale user scenario.To address the issue,this chapter transforms the sub-problem into an N players non-cooperative game,and proves that the game has Nash equilibrium.Finally,this chapter develops a distributed computation offloading strategy optimization algorithm based on the non-cooperative game.Experimental results show that the proposed distributed algorithm has better performance and flexibility than the other computation offloading strategy optimization algorithms.4.The computation offloading strategy optimization algorithms in EECCC.To study the impact of different EECCC architecture types on their service performance,according to the standard of whether cloud resources are visible to the UEs,the architecture types of EECCC are divided into two types,i.e.,hierarchical EECCC(Hi-EECCC)and horizontal EECCC(Ho-EECCC).Subsequently,this chapter studies the computation offloading strategy optimization problem in Hi-EECCC and Ho-EECCC respectively.Meanwhile,this chapter formulates the competition between UEs for ES resources as an EECCC potential game by constructing a potential function.Then,this chapter develops two computation offloading strategy optimization algorithms for the two EECCC architectures,and theoretically analyzes the convergence and performance of the algorithms in detail.The experiments study the applicability of the two EECCC architectures in detail from the time consumption of UE,the energy consumption of UE,and the resource utilization of ES.In addition,the three important conclusions summarized in this chapter can be used to guide the selection of EECCC architecture types for the different application scenario.