Comparison of Machine Learning Techniques for VNF Resource Requirements Prediction in NFV

The network function virtualization (NFV) is a developing architecture that uses virtualization technology to separate software from hardware. One of the most important challenges of NFV is the resource management of virtualized network functions (VNFs). According to the dynamic nature of the NFV, resource requirements of VNFs do not always remain static. In fact, the resource allocation to VNFs must be changed to correspond to variations of incoming traffic to the network. These changes cause a significant delay in the reallocation of resources. For this reason, applying resource estimation models before their allocation can prevent the upcoming problems and leads to performance improvement of resource allocation methods dynamically. In this paper, according to the resource prediction importance in NFV, three support vector regression (SVR), decision tree (DT) and k-nearest neighbor (KNN) algorithms of machine learning techniques are analyzed and compared. In addition, the effect of the genetic algorithm as a feature selection method on the mentioned methods is evaluated. The results show that an error less than one in SVR, DT, and KNN algorithms in predicting resources is achieved. However, the SVR algorithm has more execution time than the other two algorithms.