期刊
CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
卷 26, 期 1, 页码 613-634出版社
SPRINGER
DOI: 10.1007/s10586-022-03660-w
关键词
Edge computing; Cloud computing; Networks; Software-defined networking; Virtualized network functions; Machine learning
Emerging technologies such as network function virtualization and software-defined networking have made significant advancements in network management through software-based approaches. The combination of machine learning and cognitive techniques has introduced a knowledge plane for the Internet, enabling automation and recommendation. The proposed self-driving system based on knowledge defined networking achieves optimal path selection and resource balancing while utilizing a graph neural network for accurate estimation of key performance indicators.
Emerging technologies such as network function virtualization and software-defined networking (SDN) have made a phenomenal breakthrough in network management by introducing softwarization. The provision of assets to each virtualized network functions autonomously as well as efficiently and searching for an optimal pattern for traffic routing challenges are still under consideration. Unfortunately, the traditional methods for estimating the desired performance indicators are insufficient for a self-driven SDN. In the last decade, a combination of machine learning and cognitive techniques construct a knowledge plane (KP) for the Internet which introduces numerous benefits to networking, like automation and recommendation. Furthermore, the inclusion of KP to the conventional three planes SDN architectures recently has added another knowledge defined networking (KDN) architecture to drive an SDN autonomously. In this article, a self-driving system has been proposed based on KDN to achieve the selection of an optimal path for the deployment of service function chaining (SFC) and reactive traffic routing among the edge clouds. Considering the limited resource of edge clouds, the proposed system also maintains a balance among edge cloud resources while orchestrating SFC resources. The graph neural network has been also applied in the proposed system to recognize the composite relationship concerning topology, traffic features, and routing patterns for accurate estimation of key performance indicators. The proposed system improves resource utilization efficiency for SFC deployment by 20%, maximum network throughput by 5%, and CPU load by 13%.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据