Study Of Offloading Problems For Manufacturing Process-oriented Edge Computing Tasks

In recent years,with the proliferation of emerging computing-intensive applications based on Internet technology,people have put forward higher requirements for the connection speed and time latency of smart devices,and the current cloud computing architecture is no longer well suited for rapid processing of large-scale data.Mobile Edge Computing(MEC)reduces the data transmission distance of tasks to be processed by offloading complex tasks generated by working devices to MEC servers that are physically closer to users,thereby reducing the transmission delay of tasks and improving the processing speed of tasks.Considering that the manufacturing process of modern smart factories continuously generates huge amount of heterogeneous data,it is important to process these data in a fast and orderly way and reduce the energy consumption and load of the equipment.In this paper,we study the offloading problem of edge computing tasks in manufacturing process,firstly,we consider the offloading problem of tasks,and divide the tasks into two parts:local execution and edge offloading execution.Based on this,the tasks to be offloaded are divided into three types: split subtask parallel execution,split subtask serial execution and split subtask mixed execution,and the constraint relationship between subtasks is represented by a directed acyclic graph.For the multi-user-multi-MEC server scenario,three different task offloading methods are proposed for different types of tasks: parallel offloading,serial offloading and mixed offloading,and the corresponding mathematical model is established with the objective of minimizing the maximum task processing delay.Secondly,after the task is split and offloaded,the resource scheduling problem of MEC server needs to be considered,i.e.,how to offload the split task to the suitable MEC server to meet the optimization goal.The problem solving process is studied and analyzed using greedy algorithm,genetic algorithm and deep reinforcement learning algorithm respectively,in conjunction with the production process in manufacturing process.Based on the proposed task offloading and computing resource scheduling model,three algorithms are used to solve the edge computing task offloading problem with the objective of minimizing the maximum task processing delay.The comparative experimental analysis is carried out using multiple use cases of different scales,considering the algorithm execution time and optimization effect.The effectiveness and reliability of the algorithms in offloading methods are verified,and they can significantly reduce the latency of edge task offloading and reduce the load on MEC servers.Based on the summary of the experimental results,it is pointed out that future work can be started in the direction of multi-objective dynamic optimization.