An RBF neural network-based dynamic virtual network embedding algorithm
An RBF neural network-based dynamic virtual network embedding algorithm
复制标题
基于RBF神经网络的动态虚拟网络嵌入算法
DOI：
10.1002/cpe.4516
复制
发表时间：
2019
影响因子：
2
通讯作者：
Xue Qingshui
中科院分区：
文献类型：
--
作者：
Zheng Xiangwei;Zhang Yuang;Zhang Hui;Xue Qingshui
网络虚拟化是一种很有前途的技术，它用来解决现有Internet架构的缺点，其中虚拟网络（VN）的嵌入在资源分配中起着至关重要的作用。然而，目前的虚拟网络嵌入算法主要处理静态VN嵌入，在VN请求的生命周期内，网络资源的分配是恒定的和过多的。在现实中，用户的资源需求通常会随着时间的推移而变化，因此，静态VN嵌入算法可能会导致基板资源利用率低，运营商的收入减少。为了解决上述问题，本文重点研究了考虑动态需求的虚拟网络嵌入。我们采用监督径向基函数（RBF）神经网络预测虚拟请求的变化，并通过调整已分配的资源，动态地重新分配基板资源。仿真实验表明，与静态算法相比，基于监督RBF的动态VN嵌入算法能够更准确地预测虚拟网络的需求变化，自适应地调整资源分配和管理，提高底层网络的资源利用率。
Network virtualization is a promising technology, which is used to solve the existing shortcomings of the Internet architecture and in which virtual network (VN) embedding plays a vital role in resource allocation. However, current virtual network embedding algorithms mainly deal with static VN embedding; the allocations of network resources are constant and excessive in the VN requests' lifetime. In reality, the user's resource requirements usually change with time; hence, static VN embedding algorithms can lead to low utilization of substrate resources and reduction of the operators' revenues. To solve the aforementioned problem, we focus on the virtual network embedding considering dynamic demands in this paper. We apply supervised radial basis function (RBF) neural network to predict the changes of virtual requests and dynamically reallocate substrate resources by adjusting already allocated resources. Simulation experiments show that compared with the static algorithms, dynamic VN embedding schemes with supervised RBF can more accurately predict the demand changes of virtual networks, which adaptively adjusts resources allocation and management and therefore improves resource utilization of substrate networks.