Resource Prediction of Virtual Network Function Based on Traffic Feature Extraction

Abstract

∗ With the continuous innovation of the Internet, the development of Cloud Computing technology and standard server promotes the development of Network Function Virtualization (NFV). Although NFV solves the shortcomings of traditional network function equip-ment such as high cost and difficult operation, it also brings certain challenges. Resource management in NFV is a complex problem because the resource requirements of Virtual Network Function (VNF) vary with the dynamic traffic, so it is necessary to under-stand the resource requirements of VNF. Due to the limited physical network resources, it is very important to find an effective resource prediction method. Based on Heterogeneous Information Network (HIN) and Multilayer Perceptron (MLP), we propose VNF-RPHIN, a method of the VNF resource requirement prediction based on traffic feature extraction. Firstly, we construct the HIN by the correlation between traffic features. Secondly, we use the HIN2Vec model to obtain the feature representation of each traffic feature. Finally, the attention mechanism is used to measure the importance of each feature, and different weights are assigned to each feature, and then they are input into the MLP model. The hidden relationship between traffic features is mined by HIN to predict the resource requirement of the VNF. The experimental results show that the proposed method has good performance and is superior to the traditional machine learning model and common deep learning model.