面向工业互联网的边缘计算资源调度技术研究

With the development of industrial networks, more and more latency-sensitive applications emerge recently. These applications put more requirements on the current Internet. Edge computing is seen as an promising direction for the development of industrial networks, but it also faces many challenges, such as resource scheduling, compatibility and intelligent management..To improve the performance of edge computing for industrial networks, this project proposes a new resource schedule framework that combines the software defined technology with the serverless technology. It adopts centralized controllers to collect network states, and schedule the computing resources based on serverless technology. The new framework can reduce the latency that users experience while providing the unified interface to users, and users do not have to care about the complexities of the underlying network. To improve the performance, we also design the resource orchestration scheme based on dynamic programming and heuristic algorithms with the framework. We also investigate into the intelligent management problem based on convolutional neural network and reinforcement machine learning, to improve the network management burden and efficiency. .Research on the project could contribute to the development of theory and technology on resource scheduling technologies in edge computing for industrial network. The research can also promote the development of industrial network in the real-world. Thus, this project has significant scientific importance and widespread commercial value.

随着工业互联网快速的发展，涌现出许多延迟敏感的应用。这些应用对互联网提出了低延迟的更高要求。边缘计算被认为是促进工业互联网发展的重要方向，但也面临着资源调度、兼容性、智能管理的挑战。.本项目在工业互联网边缘计算场景下，提出设计基于软件定义与无服务化计算的轻量级资源调度框架，利用中心控制器技术收集网络状态，并基于无服务化计算技术对计算环境进行调度，使得网络在保障用户访问通信的低延迟特性的同时，向用户开放统一接口，用户不需要关心网络低层的复杂性；在框架下，为了优化用户的延迟体验，采用基于动态规划与启发式方法的计算资源编排算法；为了降低网络管理负担且提高网络效率，研究基于卷积神经网络与强化学习的智能管理机制。.本项目的研究可以为工业互联网边缘计算场景下的资源调度提供一定的理论基础和技术支持，也可以帮助推动工业互联网产业的更快发展，因此，本项目的研究具有重要的科学意义和应用价值。

设计和实现了面向工业互联网的边缘计算技术，基于端云间的资源调度实现端云间的低延迟和高可靠选路。在边缘场景包括新型5G工业互联网场景，提出了基于机器学习的资源调度与编排算法，满足用户QoE需求的同时，提高资源利用率，实现面向计算、存储与网络的一体化资源编排与管理方案。所设计的边缘网络已在中国移动的网络中部署，共纳管了数十个真实的边缘计算节点，相关成果发表于软件学报、IEEE IoT等期刊。与传统的以云计算为中心的计算平台相比，在节约计算成本的同时降低了37%的计算延迟，相关成果发表于IEEE IoT、IEEE Transactions on Industrial Informatics等期刊。基于开源系统Openwhisk搭建了基于无服务化计算的原型系统，并与中国移动合作部署了十余个节点。发表SCI论文10篇，申请发明专利3项。