Research On Task Offloading And Resource Optimization In Edge Computing Systems

With the development of wireless communication technologies and software technologies,a large number of new intelligent applications and services,such as virtual reality/augmented reality(VR/AR)have emerged,put forward higher requirements for delay and processing performance.Simply optimize the data transmission cannot meet the users' service requirements of new services.The traditional cloud computing system has sufficient computing resources,but it will inevitably cause a large round-trip delay because it is far from the user.The emergence of edge computing system provides the possibility to solve the above problems.By sinking the functions of cloud computing to the edge of the network,edge computing supports efficient data transmission and caching services,and provides sufficient computing capabilities which can significantly improve the latency and energy consumption performance.However,with the limitation of multi-dimensional resources,including computing,caching,and transmission,and in the face of the trend of large-scale device connection in the future network and service performance requirements,the integration and reasonable allocation of limited multi-dimensional resources will become increasingly important.The intensive deployment of cells makes the handover of services more frequent,which easily leads to the failure of task offloading;the heterogeneity and imbalance of multi-dimensional resources restrict the performance improvement of edge computing system at the same time;the heterogeneity and proximity of devices in Device-to-Device(D2D)network make D2D-based distributed offloading an important way to relieve the offloading pressure.Considering the challenge,this paper studies the task offloading and resource allocation problem in edge computing systems from three aspects:user mobility analysis,cache-computing-transmission integration,and multi-mode collaborative offloading enhanced with distributed computing of devices.The main work of this disstertation is as follows:1.The problem of task offloading and resource allocation based on users' mobility analysis is studied.Firstly,users are divided into strong-mobile users and weak-mobile users based on their mobility,and an optimization problem is constructed to maximize users' revenue.The problem is a hybrid discrete non-convex optimization problem.Specially,in this problem,we jointly consider the user access allocation strategy,the power allocation strategy of the full-duplex base stations,the user offload strategy,and the computing resource allocation strategy.Secondly,considering that the mobile users may handover between cells which causing task processing failures,the exponential distribution is used to model the user's sojourn time.Based on the distribution of user's sojourn staying time,the optimization strategies of computing resources and the offloading proportion are designed,ensuring the processing results can be successfully sent back to the users before they leave the coverage of their severing cells.We determined the optimal offloading ratio for such types of users,which transformed the optimization problem into a linear optimization problem(LP),which can be solved quickly.Furthermore,for weak-mobile users,considering the non-convexity and complexity of the optimization problem,we decompose the problem into multiple sub-problems.The optimal unloading ratio for the original optimization problem and the convexity of the resource optimization problem are proved.The dual Lagrangian method is introduced to solve the resource optimization problem,by which the optimal power allocation policy and the resource allocation policy are obtained.Then the resource optimization problem is combined with the user allocation problem and solved by a greedy-based joint optimization algorithm.The simulation results show that the proposed algorithm has better performance through comparison with various algorithms.2.The problem of task offloading and resource allocation in cache-enhanced edge systems is studied.Firstly,according to the characteristics of users' requests and traffic,we discussed the cache placement problem,the task offloading problem in two scenarios.For the cache placement scenario,the stochastic theory is used to model the offloading delay and energy consumption with the random task caching of users under the Poisson distribution,and the utility maximization problem is constructed.Due to the non-convexity of the cache placement optimization problem,a genetic-based random caching algorithm was proposed to obtain a sub-optimal solution.Secondly,we model the task offloading and resource optimization problems according to the utility function of the task offloading scenario.In this problem,we considered the joint optimization of user offloading strategy,computing resource allocation strategy,and offloading mode selection strategy.As the original problem is non-convex,we break it down into three sub-problems:user access problem,offloading mode selection problem,and resource allocation problem.Finally,we propose a user access algorithm based on the revenue function,and a resource allocation algorithm based on the Lagrangian duality.Using game theory,we proposed an iterative joint optimization algorithm to obtain a global solution.Through a large number of simulation results,the proposed performance is verified through comparison with various algorithms.3.The problem of multi-mode collaborative task offloading and resource allocation based on D2D distributed computing is studied.Firstly,a multi-user edge computing framework,including multi-dimensional resource integration is considered,and four collaborative computing models are proposed.The framework can schedule four task offload modes according to the attributes of user offloading requirements,to improve the utilization of the system.Secondly,a cost function of the system with a trade-off of servicing delay and energy consumption is constructed,and an optimization problem is proposed to minimize the system service cost.In the optimization problem,the joint optimization of the multi-cooperative mode selection strategy and the offload task proportion allocation,computing resources,and cache resources are considered.Considering that the joint optimization problem is a multi-variable,high-dimensional 01 mixed non-convex optimization problem,the original problem is segmented by simplified problem decomposition.Thirdly,for the four cooperative offloading modes,the optimal offloading task ratio and resource allocation scheme under each mode are obtained through theoretical analysis.With the adoption of game theory,the optimal solution of the joint optimization problem is obtained with iterative methods.Through simulations,this scheme is compared with other related schemes to illustrate its performance,and the experimental results are analyzed and discussed at last.