A Technical Research Of Destination Driven Computation Offloading For Edge Computing

With the exponential growth of the number of intelligent terminal devices in the In-ternet of things,the technology of Computation Offloading is widely used in IoT Edge Computing to carry out a large number of complex computation tasks.According to the target of the calculation result,the computation offloading can be divided into two modes:source-driven computation offloading and destination-driven computation offloading.Most of the current research focuses on the former,while there is very little research on the lat-ter.Therefore,there is a need for a technical solution that is suitable for destination-driven computation offloading scenario.First,this paper summarizes the current scenarios of source-driven computation of-floading and the research at home and abroad.Then,it introduces the requirements and re-search significance of destination-driven computation offloading scenarios.In the destination-driven computation offloading system architecture,a strategy of destination-driven step-by-step computation offoading along the route(DDSCOR)is proposed for delay-sensitive application services.Combining with data transmission and computation offloading,DDSCOR reduces the overall delay from the source node releasing the task to the results arriving the destination node.This paper further analyzes the advantages and innovations of DDSCOR computation offloading technology compared to traditional computation offloading tech-nology.second,in the destination-driven computation offloading scenario,the computation offloading strategy can be divided into complete information and incomplete information according to the different information interaction method of the computing servers.This paper mainly focuses on the specific implementation strategies of DDSCOR computation offloading in two different network:(1)Under the scenario of complete information network with global interactive per-ception of the computing servers,a static computation offloading model is proposed for the overall realization from the source node to the destination node,and the optimal DDSCOR offloading strategy is transformed into a solution with the optimal goal of minimizing the overall delay.After proving that the optimization problem is an NPC problem,an im-proved Ant Colony Algorithm is proposed to obtain the approximate solution.Simulation results show that the algorithm can quickly obtain an accurate minimum time delay and the offloading path of DDSCOR,which can achieve the optimal solution of the execution delay compared with the traditional computation offloading strategy.(2)Under the scenario of incomplete information network with local interactive per-ception of the computing server,a dynamic computing offloading mode is proposed for each step in the process from the source node to the destination node.The DDSCOR optimal computation offloading strategy is decomposed into each computing server par-ticipating in the step-by-step computation offoading along the route,and make the next best execution action according to the current environment,so as to achieve the minimum delay from the source node to the target node.Due to the changes of the parameters of the optimization problem,this paper proposes the DQN(deep-q-learning)algorithm of Reinforcement Learning to solve it.Simulation results show that the DDSCOR dynamic offloading strategy can reduce the overall execution delay quickly and effectively.