Research On Task Offloading Methods For Multi-Access Edge Computing Networks

With the widespread deployment of fifth-generation mobile communication networks and the rapid advancements in sixth-generation mobile communication technologies,intelligence and computing technologies have become integral to communication networks.This evolution has given rise to new types of services and applications,including extended reality,digital twins,intelligent transportation,intelligent energy,intelligent manufacturing,and the intelligent Internet of Things.Consequently,there has been an explosive increase in the number of terminals,data traffic,and demand for computational power.Multi-access edge computing(MEC)and MEC networks(MECNs)have emerged to address the increasingly complex and diverse communication,computing,and intelligence requirements across various industries and applications.MECNs deploy computing power,storage resources,and service environments to the edge of multiple access networks,facilitating the execution of applications and computing tasks in proximity to the client.This proximity offers advantages such as real-time processing,enhanced reliability,heightened security,scalability,and location awareness.Task offloading stands out as a pivotal technology within MECNs.It effectively supplements the computing power,storage capacity,and energy efficiency shortcomings of terminal devices.This acceleration in computing task processing not only boosts performance but also prolongs the standby time of terminal devices.Thus,MECNs and task offloading methods are recognized as indispensable technologies propelling the advancement of sixth-generation mobile communication networks.Current research on task offloading for MECNs is extensive,yet it presents several challenges.Relevant MECNs research illustrates the feasibility of using MEC technology for wireless Mesh networks,but there is a lack of specific research on task offloading.Moreover,most MECNs research is oriented to wireless access networks and fails to jointly utilize the resources of wireless and wireline access networks.Traditional task offloading methods face limitations such as local optimization,lack of generalization,and inability to adapt to dynamic environments when dealing with complex,large-scale,and highly dynamic task offloading problems across heterogeneous access networks.Additionally,most of the related LEO satellite MECNs research directly assumes a fixed number of satellites or satellite orbits,ignoring the offloading destination satellite selection problem caused by the movement of LEO satellite constellations.To address these challenges,this thesis focuses on two main aspects: modeling MECNs architecture and designing task offloading algorithms.This thesis conducts in-depth research on the task offloading methods for MECNs.The main research efforts and innovative contributions are as follows:(1)Aiming at the current situation that the research on task offloading for wireless Mesh networks has not been carried out yet,this thesis investigates the task offloading algorithm for wireless Mesh MECN.Establish a wireless Mesh MECN architecture,develop a task offloading model,propose the area-based offloading policy(ABOP)algorithm,and enable task offloading from Mesh client to either Mesh gateway or Mesh routers.Simulation results demonstrate that the ABOP algorithm can reduce the average task delay and enhance the number of successfully executed tasks.(2)Aiming at the problem that existing research on task offloading in MECNs does not jointly utilize the resources of wireless wireline access networks,this thesis investigates the task offloading algorithm for wireless wireline MECN.Establish a wireless wireline MECN architecture,develop a task offloading model from mobile device to wired device,propose the guard time-based computation offloading(GTCO)algorithm,and enable controllable task offloading with an adjustable percentage of different computation methods.Simulation results indicate that the GTCO algorithm can enhance computing resource utilization,decrease task failure rate,and reduce average task delay.(3)Aiming at the problem of intelligent decision-making for task offloading across heterogeneous access networks,this thesis investigates the task offloading algorithm for wireless wireline MECN.Enhance the architecture of the wireless wireline MECN and develop models for wireless wireline communication,task offloading and shunting,as well as task queuing.Derive the average total delay of tasks which includes wireless and wireline network delay,task queuing delay,and task computing delay.The optimization problem is formulated as minimizing the average total delay of the task and transformed into a deep reinforcement learning(DRL)problem using Markov decision processes,and a DRL-based computation offloading and shunting(DRLCOS)algorithm is proposed.Simulation results demonstrate the ability of the DRLCOS algorithm to enhance computational resource utilization rates and reduce the average total delay of tasks.(4)Aiming at the problem of offloading destination satellite selection due to the movement of LEO satellite constellations,this thesis researches the task offloading algorithm for LEO satellite MECN.Establish the architecture of the LEO satellite MECN and develop a communication and computation model,as well as a task offloading and queuing model.Derive the average total delay of tasks,including transmission delay,propagation delay,queuing delay,and computation delay.Formulate a nonlinear integer programming(NLIP)problem to minimize the average total delay of tasks.Considering the mobility of low Earth orbit satellite constellations,decompose the NLIP problem into access satellite selection and neighboring satellites selection sub-problems.Propose the LEO satellites selection-based computation offloading(LSSBCO)algorithm.Simulation results demonstrate the capability of the LSSBCO algorithm to minimize the average total delay of tasks under the constraint of computational resource utilization.The MECNs architecture investigated in this thesis and the proposed task offloading methods can provide theoretical support and technical guidance for subsequent research.