| 非稳态雾赋能网络中的在线任务卸载方法 |
| |
| 作者姓名： | 朱兆伟 刘婷 钱骅 罗喜良 |
| |
| 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. 中国科学院大学, 北京 100049;
3. 上海科技大学, 上海 201210;
4. 中国科学院上海高等研究院, 上海 201210 |
| |
| 基金项目： | 国家自然科学基金（61671436）资助 |
| |
| 摘 要： | 为充分发掘分布在不同位置上的雾节点的计算资源，任务卸载被寄予众望。在雾计算场景下，以尽可能减少任务卸载的长期成本为目标，试图寻找一个高效的在线任务卸载方法。为此，这一问题被建模成一个随机规划问题，该问题中系统参数所对应的随机变量的期望会在未知时刻突变，系统参数相关信息只能在任务完成后的反馈中获得。基于非稳态多臂老虎机模型，提出一个高效的算法来解决这一具有挑战性的随机优化问题，给出理论分析证明该算法的渐进最优性。数值实验证明了该算法的优越性。
|
| 关 键 词： | 在线学习 雾计算 任务卸载 随机优化 多臂老虎机 |
| 收稿时间： | 2019-02-15 |
| 修稿时间： | 2019-03-28 |
Online task offloading in non-stationary fog-enabled networks |
| |
| Authors: | ZHU Zhaowei LIU Ting QIAN Hua LUO Xiliang |
| |
| Institution: | 1. Shanghai Institute of Microsystem&Information Technology, Chinese Academy of Sciences, Shanghai 200050, China;
2. University of Chinese Academy of Sciences, Beijing 100049, China;
3. ShanghaiTech University, Shanghai 201210, China;
4. Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China |
| |
| Abstract: | To fully exploit the computational resources in different fog nodes, task offloading is emerging. In this work, under the fog computing scenario, an efficient online task offloading strategy is investigated to minimize the long-term cost of task offloading. To achieve this goal, the problem is modeled as a stochastic optimization problem. Moreover, the system parameters are characterized by random variables, and their expectations may change abruptly at unknown time slot. Besides, the information about the system parameters is only available through the feedbacks after the task finishes. Using the non-stationary multi-armed bandit framework, we propose an efficient algorithm to handle this challenging stochastic programming. Furthermore, theoretical analyses are presented to prove the asymptotic optimality of the proposed algorithm. Numerical results reveal the advantages of this algorithm. |
| |
| Keywords: | online learning fog computing task offloading stochastic optimization 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 |
|
