| 动态雾计算网络中基于在线学习的任务卸载算法 |
| |
| 作者姓名： | 谭友钰 陈蕾 周明拓 王昆仑 杨旸 张武雄 |
| |
| 作者单位： | 1. 中国科学院上海微系统与信息技术研究所, 上海 200050;
2. 国网浙江省电力有限公司, 杭州 310007;
3. 中国科学院大学, 北京 100049;
4. 上海科技大学, 上海 201210;
5. 上海雾计算实验室, 上海 201210 |
| |
| 基金项目： | 国家电网科技项目（52110418001U）和国家自然科学基金（61571004）资助 |
| |
| 摘 要： | 任务卸载是雾计算的主要技术之一，即计算能力不足的节点将任务卸载给具有富余资源的节点帮助计算。以优化任务平均卸载时延和提升卸载服务成功率为目标，利用多臂老虎机理论为动态雾计算网络提出一种基于在线学习的任务卸载算法，可实时做出最优卸载决策。将该算法扩展到非稳定网络状态，使之可以动态追踪网络中节点的资源与环境变化，实时调整卸载决策。详细分析所提出算法的性能、复杂度和存储占用情况。仿真结果表明，这两种算法可达到的长期平均任务卸载时延均十分接近理想算法下的最优时延，卸载服务成功率也得到显著提升。此外，所提算法在非稳定的网络状态下能够追踪到计算资源与环境的变化。
|
| 关 键 词： | 雾计算 任务卸载 在线学习 多臂老虎机 |
| 收稿时间： | 2019-01-30 |
| 修稿时间： | 2019-03-28 |
Online learning-based task offloading algorithms for dynamic fog networks |
| |
| Authors: | TAN Youyu CHEN Lei ZHOU Mingtuo WANG Kunlun YANG Yang ZHANG Wuxiong |
| |
| Institution: | 1. Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China;
2. State Grid Zhejiang Power Co Ltd, Hangzhou 310007, China;
3. University of Chinese Academy of Sciences, Beijing 100049, China;
4. ShanghaiTech University, Shanghai 201210, China;
5. Shanghai Institute of Fog Computing Technology, Shanghai 201210, China |
| |
| Abstract: | Task offloading is one of the main techniques of the fog computing, and it means that the computation-limited nodes offload the tasks to the capable nodes for help. Firstly, we propose TOD (online learning-based task offloading algorithm for the dynamic fog networks under stationary status) using the MAB (multi-armed bandit) theory, which aims at minimizing the long term offloading delay and improving the task offloading success ratio. Then, we propose TOD-N (online learning-based task offloading algorithm for the dynamic fog networks under non-stationary status) to efficiently track the changes of the sharing computing resources and the channel environment. Moreover, we analyze the performances of the two algorithms on the optimality, the computational complexity, and the memory usage. Simulation results show that the long term average offloading delays achieved by the two algorithms are almost as good as the one achieved by the Oracle algorithm, and the offloading success ratios are also efficiently promoted. Moreover, TOD-N tracks the optimal resources efficiently under non-stationary network status. |
| |
| Keywords: | fog computing task offloading online learning multi-armed bandit |
|
| 点击此处可从《》浏览原始摘要信息 |
| 点击此处可从《》下载免费的PDF全文 |
|
