Research On Virtual Network Mapping Algorithm For Dynamic Resource Demand

Network virtualization technology has effectively solved many shortcomings of the existing network architecture,which has become one of the mainstream technologies of the next generation Internet development.One of the key technologies of network virtualization is to manage and utilize the network resources reasonably and effectively,and the problem of virtual network embedding is the core of network resource allocation in network virtualization.The problem of virtual network embedding can be divided into static embedding problem and dynamic embedding problem.The current researches of virtual network embedding are mostly limited to the traditional static virtual network embedding methods,where the constraints of virtual requests are previously known and the resources assigned to the virtual network are invariable until the end of their lifetime.But in actual life,because the network load is changing with the change of the user’s demand and the traditional resource allocation methods are based on the peak of the virtual network request to allocate the fixed number of resources,which will lead to the low utilization of the whole substrate network resources.Therefore,how to share the substrate resources reasonably and effectively has become the focus of network virtualization research.In addition,the reconfiguration algorithms are mostly used in the current research of dynamic virtual network embedding,which only consider the redistribution of substrate resources,but do not take into account the dynamic changes of the user’s request.In recent years,machine learning based intelligent algorithms have been widely used to solve various resource allocation problems.Based on the above reasons,the thesis aims at strengthening the distribution and management of the underlying substrate network resources,and studies the virtual network embedding algorithm with dynamic change requirements.The specific research contents include:⑴ A dynamic virtual network embedding algorithm(SVR-VNE)based on support vector machine is proposed.The principle of support vector machine is studied,and support vector machine regression prediction is realized.In order to improve the precision of support vector machine regression model,the PSO algorithm is used to optimize the parameters of the SVR model.The experiment shows that the optimized SVR-VNE algorithm can capture the change of virtual request more accurately,which has achieved good results in the acceptance rate,income and cost,and improved the utilization of network resources.⑵ A dynamic virtual network embedding algorithm(RBF-VNE)based on RBF neural network is proposed,which has good nonlinear fitting ability,simple training and fast convergence.Experiments show that the RBF-VNE algorithm not only effectively utilizes network resources,but also increases the revenue of network operators.⑶ On the basis of SRBF regression prediction,a dynamic virtual network embedding algorithm(SRBF-VNE)based on supervised training RBF is proposed by combining the supervised training method with the RBF algorithm.Experiments show that this method can effectively manage the underlying physical network resources without changing the mapping cost.