ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展

• •    下一篇

基于资源需求预测的动态服务功能链迁移方法

阳勇,孟相如,康巧燕,陈港   

  1. (空军工程大学信息与导航学院 西安 710077)(yangyong_123456789@163.com)
  • 出版日期: 2022-08-24

Dynamic Service Function Chain Migration Method Based on Resource Requirements Prediction

Yang Yong, Meng Xiangru, Kang Qiaoyan, Chen Gang   

  1. (Information and Navigation School, Air Force Engineering University, Xi’an 710077)
  • Online: 2022-08-24

摘要: 针对网络功能虚拟化环境下服务功能链资源需求变化引起的底层网络过载问题,提出一种基于资源需求预测的动态服务功能链迁移方法.首先,综合考虑迁移开销和迁移后底层网络的资源占用情况,建立底层网络开销模型.其次,利用经验模态分解将资源需求序列分解成本征模函数分量与残差分量,再通过径向基函数神经网络实现对各分量的预测,神经网络的训练过程采用粒子群算法进行参数优化.最后,对下一时隙即将过载的物理节点或链路,选择对过载资源占用最多的虚拟网络功能或虚拟链路进行迁出,并基于流量优化的原则,通过对全局拓扑的感知选择能最小化底层网络开销的物理节点迁入.仿真实验表明,所提的资源需求预测模型在提高预测精度的同时能缩短预测时间,所提的服务功能链迁移方法在降低底层网络开销、减少端到端时延和提高服务功能链可靠性等方面有较好性能.

关键词: 网络功能虚拟化, 服务功能链, 虚拟网络功能, 资源需求预测, 径向基函数

Abstract: Aiming at the problem of network overload caused by the change of resource requirements of service function chain (SFC) under network function virtualization (NFV) environment, a dynamic SFC migration method based on resource requirements prediction (RRP-DSFCM) was proposed. Firstly, the migration overhead and resource overhead are considered comprehensively, and the physical network overhead model is established. Secondly, the resource requirements sequence was decomposed into intrinsic mode function (IMF) component and residual component by empirical mode decomposition (EMD), and each component was predicted by radial basis function (RBF) neural network. Particle swarm optimization (PSO) algorithm was used in the training process of neural network to optimize the parameters. Finally, for the physical nodes or links that will be overloaded in next timeslot, the virtual network function (VNF) or virtual link that takes up the most overload resources are selected to move out, and based on the principle of traffic optimization, the physical node which can minimize the overhead of the physical network are selected to move in through the awareness of the global network topology. The simulation results show that the proposed resource requirements prediction model can shorten the prediction time while improving the prediction accuracy, and the proposed SFC migration method has good performance in reducing the physical network overhead, reducing the end-to-end delay and improving the reliability of the SFC.

Key words: network function virtualization, service function chain, virtual network function, resource requirements prediction, radial basis function

中图分类号: