Research On Edge Computing Task Offloading Method Based On PID Control

While the Internet of Things is realizing the interconnection of everything,the influx of massive data has put tremendous pressure on the core data network,and with the emergence of low-latency application scenarios such as intelligent monitoring and autonomous driving,transferring data to remote cloud service centers causes great transmission delays and cannot meet the demand for real-time.Edge computing,as an extension of cloud computing,shifts some tasks to the edge for execution through task offloading,reducing the amount of data flowing to the cloud center while lowering the data transmission latency.While edge devices have varying performance and limited resources,how to reasonably perform task offloading scheduling is an very important research direction.Existing edge task offloading approaches commonly solve the offloading decision problem by extensive computation and constraint modeling,but the performance of end nodes is weak and cannot provide sufficient computational resources to support complex offloading strategies.To address these problems,an adaptive offloading scheduling method based on Proportional Integral Derivative(PID)controller is proposed based on the analysis of several typical edge task offloading methods,whose core is to use Directed Acyclic Graph(DAG)to simulate the tasks with dependencies,to represent the data dependencies by the direction of the edges in the DAG,and to build a closed-loop PID feedback regulation system to guide the offloading process.When the performance of the system fluctuates,it can predict the gain value after offloading the task with low computational overhead,quickly adapt to the new system environment,and achieve a stable offloading effect.To verify the effectiveness of the proposed algorithm,experimental comparisons are conducted using the i Fog Sim simulator around typical application scenarios including control systems for vehicles in autonomous driving,safety monitoring systems in smart transportation,and video streaming service systems.The experimental results show that the maximum end-to-end execution delay reduction of 43.3% and the total energy consumption reduction of 7.5% can be achieved in the aforementioned application scenarios compared with two representative algorithms.