Application-aware QoS routing in SDNs using machine learning techniques
Application-aware QoS routing in SDNs using machine learning techniques
复制标题
DOI：
10.1007/s12083-021-01262-8
复制
发表时间：
2021-11
影响因子：
4.2
通讯作者：
Weichang Zheng;Mingcong Yang;Chenxiao Zhang;Yu Zheng;Yunyi Wu;Yongbing Zhang;Jie Li
中科院分区：
文献类型：
--
作者：
Weichang Zheng;Mingcong Yang;Chenxiao Zhang;Yu Zheng;Yunyi Wu;Yongbing Zhang;Jie Li
软件定义网络已经成为克服传统网络局限性的有效和有前途的手段，例如，通过保证各种应用的相应服务质量（QoS）。与传统网络固有的分布式特性相比，SDN在逻辑上是集中式的，可以利用机器学习技术来跟踪每个应用的传输需求。在本研究中，我们首先发展一个有效的数据降维方法，考虑数据项之间的相关系数。我们将交通数据分为区分类别的QoS要求的基础上，通过监督机器学习方法。然后，我们提出了一个QoS感知路由（QAR）算法，根据每个应用程序的QoS要求，找到一个路径的最小平均链路占用时间或最大平均路径剩余容量。机器学习模型的准确性表明，我们提出的降维方法比其他数据预处理方法更有效，阻塞概率的结果表明，我们的QAR算法明显优于以前的算法。
Software Defined Networking has become an efficient and promising means for overcoming the limitations of traditional networks, e.g., by guaranteeing the corresponding Quality of Service (QoS) of various applications. Compared with the inherent distributed characteristics of the traditional network, SDN is logically centralized and can utilize machine learning techniques to keep track of transmission requirements of each application. In this research, we first develop an efficient data dimension reduction approach by considering the correlation coefficients between data items. We classify the traffic data into distinguished categories based on the QoS requirements by a supervised machine learning method. Then, we propose a QoS Aware Routing (QAR) algorithm according to the QoS requirements of each application that finds a path with either the minimum average link occupied times or the maximum average path residual capacity. The accuracy of machine learning model shows that our proposed dimension reduction approach is more effective than other data preprocessing methods, and the results of blocking probability indicate that our QAR algorithm outperforms significantly previous algorithms.