区块链赋能物联网中联合资源分配与控制的智能计算迁移研究  

Resource Allocation and Control Co-aware Smart Computation Offloading for Blockchain-Enabled IoT

作　　者：陈思光[1] 王倩[1] 张海君 王堃 CHEN Si-Guang;WANG Qian;ZHANG Hai-Jun;WANG Kun(Jiangsu Key Lab of Broadband Wireless Communication and Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003;Department of Communication Engineering,University of Science and Technology Beijing,Beijing 100083;Department of Electrical and Computer Engineering,University of California Los Angeles,Los Angeles CA90095 USA)

机构地区：[1]南京邮电大学江苏省宽带无线通信和物联网重点实验室,南京210003 [2]北京科技大学通信工程系,北京100083 [3]加州大学洛杉矶分校电子与计算机工程系,洛杉矶CA90095美国

出　　处：《计算机学报》2022年第3期472-484,共13页Chinese Journal of Computers

基　　金：国家自然科学基金(61971235,61771258);江苏省“六大人才高峰”高层次人才项目(XYDXXJS-044);江苏省“333高层次人才培养工程”;南京邮电大学‘1311’人才计划;中国博士后科学基金(面上一等)(2018M630590);江苏省博士后科研资助计划(2021K501C);赛尔网络下一代互联网技术创新项目(NGII20190702)资助

摘　　要：大数据场景下,远程云服务器通常被部署用于数据处理与价值挖掘,但在面对时延敏感型或需要动态频繁交互的业务时,该种处理模式显得力不从心.作为对云计算模式的补充,雾计算因其可有效降低任务处理时延、能耗与带宽消耗而备受关注;同时,面向雾计算的计算迁移机制因其能有效缓解节点的处理负担并改善用户体验而成为领域研究焦点.在雾计算模式下,为了更好地满足计算密集型任务对时延与能耗的要求,基于区块链赋能物联网场景,本文提出了一种联合资源分配与控制的智能计算迁移方案.具体地,规划了一个在时延、能耗与资源约束下的最小化所有任务完成总成本的优化问题,其总成本构成综合考量了时延、能耗和挖掘成本,通过对通信、计算资源与迁移决策的联合优化,实现总成本的最小化.为完成任务迁移,终端以矿工的身份向雾节点挖掘(租借)计算资源,所提出的基于区块链技术的激励机制可充分调动终端和雾节点参与计算迁移的积极性并保障交易过程的安全性,设计的奖励分配规则可保证成功挖掘资源终端收获奖励的公平性.为解决上述规划的优化问题(即混合整数非线性规划问题),提出了一个联合通信、计算与控制的智能计算迁移算法,该算法融合深度确定性策略梯度算法思想,设计了基于反梯度更新的双“行动者-评论家”神经网络结构,使训练过程更加稳定并易于收敛;同时,通过对连读动作输出进行概率离散化运算,使其更加适用于混合整数非线性规划问题的求解.最后,仿真结果表明本文方案能以较快的速度收敛,且与其他三种基准方案相比,本文方案的总成本最低,例如,与其中性能最好的基于深度Q学习网络的计算迁移方案相比,总成本平均可降低15.2%.Under the scenario of big data,the remote cloud server is usually deployed for data processing and value mining,but in the face of delay-sensitive or dynamic and frequent update applications,this processing paradigm appears to be inadequate from reality.As a complement to cloud computing paradigm,fog computing has attracted great attention for which can effectively reduce task processing delay,energy and bandwidth consumptions.At the same time,fog computing based computation offloading mechanism has become a research focus,because it can effectively alleviate the processing burden of nodes and improve the experience of users.Under the fog computing paradigm,in order to better meet the requirements of delay and energy consumption for computation-intensive tasks,based on the blockchain-enabled Internet of Things(IoT)scenario,this paper proposes a resource allocation and control co-aware smart computation offloading scheme.Specifically,an optimization problem is formulated to minimize the total cost of all tasks under the constraints of delay,energy consumption,and communication and computation resources.Based on the comprehensive consideration,the component of the total cost includes the delay,energy consumption and resource mining costs;it can achieve the minimization of the total cost by jointly optimizing the communication resource,computation resource and offloading decision.In order to inspire the active participation of terminals and the fog node in the computation offloading process,and more close to the needs of the real scenario,an incentive mechanism is designed in this paper.That is,to complete the offloading of tasks,the terminal mines(rents)computation resource from the fog node as a miner and the fog node charges a certain fee according to the needed resource of terminal.For the terminals that successfully obtain the resources to complete the tasks efficiently,the system will allocate the corresponding rewards according to the occupation ratios of gained computation resource,which ensures the fairness

关 键 词：计算迁移 雾计算 区块链 深度强化学习 资源分配 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]