A task unloading strategy of IoT devices using deep reinforcement learning based on mobile cloud computing environment
- PDF / 837,158 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 95 Downloads / 169 Views
(0123456789().,-volV)(0123456789(). ,- volV)
A task unloading strategy of IoT devices using deep reinforcement learning based on mobile cloud computing environment Hui Qi1 • Xiaofang Mu1 • Ying Shi1 Accepted: 24 September 2020 Ó Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Aiming at the task unloading mode in cloud computing environment, the task unloading problem for IoT devices is studied. Through theoretical analysis, we can know that in the task unloading problem, it is usually contradictory to improve the utilization of cloud resources and reduce the task delay. In order to solve this problem, a task unloading scheme for Internet of things devices using deep reinforcement learning algorithm is proposed. The deep reinforcement learning algorithm is used to model the task unloading problem. The return value with weight is introduced into the algorithm, and the utilization rate of cloud resources and the delay of unloading task are weighed by adjusting the return value of the weight. First of all, the improved k-means clustering algorithm with weighted density is used to cluster the physical machines. The physical machines of each cluster have similar bandwidth and task waiting time. Then, deep reinforcement learning is used to select the best physical machine cluster from the current unloading tasks. Finally, the improved PSO algorithm is used to select the optimal physical machine from the optimal cluster, and Pareto is used to improve the convergence speed. Experimental results show that compared with the traditional method, the proposed algorithm has a good performance, and can achieve the goal of increasing the utilization of physical machine resources and reducing task delay. Keywords Task unloading Deep reinforcement learning Mobile cloud computing K-means clustering algorithm Physical machine Convergence speed
1 Introduction There are many tasks in mobile devices that need a lot of computing resources and consume a lot of energy. Unloading these tasks to the remote cloud computing center can effectively reduce the energy consumption of mobile devices and achieve the purpose of expanding the capacity of mobile devices [1–3]. In the related research, many scholars have proposed many different methods of task unloading. The factors considered in these task unloading methods include: energy consumption of mobile devices, network communication bandwidth, delay, capacity of cloud servers, etc.[4, 5]. By considering these factors, the existing task unloading strategy will compare the total cost of executing in the mobile device with that in & Hui Qi [email protected] 1
Department of Computing, Taiyuan Normal University, Shanxi 030619, Jingzhong, China
the cloud, so as to decide whether to unload the task to the cloud [6, 7]. Cloud computing provides a technical basis for task unloading in mobile cloud computing environment. One of the core technologies of cloud computing is virtualization, through which the physical machine in the cloud can run multiple operating sys
Data Loading...
