Intelligent Latency-Aware Virtual Network Embedding for Industrial Wireless Networks
Intelligent Latency-Aware Virtual Network Embedding for Industrial Wireless Networks
复制标题
DOI：
10.1109/jiot.2019.2900855
复制
发表时间：
2019-10-01
影响因子：
10.6
通讯作者：
Guan, Xinping
中科院分区：
文献类型：
--
作者：
Li, Mingyan;Chen, Cailian;Guan, Xinping
工业无线网络（IWN）的日益普及是由具有严格时效性要求的各种应用驱动的。然而，传统工业网络的一应用一网络架构中根深蒂固的僵化，阻碍了工业网络向智能工厂的演进。作为一种解决方案，基于切片的网络虚拟化（NV）打破了应用程序和网络基础设施之间的紧密耦合，从而提供了一个更灵活和可扩展的IWN架构。NV的应用依赖于在底层基础设施上实例化多个虚拟网络（VN）的算法，称为VN嵌入（VNE）。然而，现有的VNE算法不一定是最佳的IWNs由于缺乏QoS兼容的能力。为此，所谓的iVNE，智能延迟感知VNE方案，提出了各种工业VN（IVN），其中涉及静态嵌入和动态转发提供最后期限保证。在静态阶段，针对新到达的IVN引入了一种无路径嵌入算法，以粗粒度满足其资源需求和截止时间。然后，将动态anypath转发方法并入iVNE中，以通过深度Q学习提供智能延迟感知，从而可以及时进行转发调整，以应对链路质量和网络工作负载的动态变化。仿真结果验证了该算法的学习效率以及在动态环境下通过响应式转发实现负载均衡的能力。
The growing popularity of industrial wireless networks (IWNs) is driven by various applications with stringent timeliness requests. However, the ossification, deep-rooted in the one-application one-network architecture of traditional IWNs, impedes the evolution of IWNs toward smart factory. As a solution, the slice-based network virtualization (NV) breaks the tight coupling between applications and network infrastructure, and thus provides a more flexible and scalable IWN architecture. The application of NV relies on the algorithms that instantiate multiple virtual networks (VNs) on a substrate infrastructure, known as VN embedding (VNE). However, existing VNE algorithms are not necessarily optimal for IWNs due to the absence of QoS-compliant capacity. To this end, so called iVNE, an intelligent latency-aware VNE scheme, is proposed to provide deadline guarantee for various industrial VNs (IVNs), which involves both static embedding and dynamic forwarding. In the static stage, an anypath embedding algorithm is introduced for the new arrival of IVNs so that their resource demands and deadlines can be satisfied with coarse grain. Then, a dynamic anypath forwarding method is incorporated into iVNE to offer intelligent latency sensing via deep Q-learning, and thus forwarding adjustments can be made timely to address the dynamic changes of link quality and network workload. The simulation results are provided to demonstrate the learning efficiency as well as the ability of load-balancing through responsive forwarding under dynamic environment.