Research On Virtual Network Embedding Algorithm For The Google Cloud Dataset

As one of the basic features in the future network,network virtualization separates the two functions of network service and infrastructure,and provides appropriate network services to users according to their needs.One of the key problems in network virtualization is the virtual network embedding,which is to allocate the underlying network resources for a virtual network request having nodes and link resource constraints.If we can utilize the infrastructure resources with optimization,it can bring great benefits to the infrastructure providers.Thus the problem of virtual network embedding has been widely concerned in recent years.However,most of the current researches on the algorithm of virtual network embedding generate network topology models using the tools of GT-ITM for simulating the network environment,and they do not consider the complexity of the real network environment.Therefore,these researches cannot give authoritative conclusions to the algorithm choice in different scale network environments.Besides,the existing virtual network embedding algorithms all have some limitations.When a large number of requests arrive in batch form in the network environment,the embedding performance of the virtual network will be significantly reduced if not considering the problems mentioned above,thus the benefit of the underlying resources is also reduced.Based on the above points,the research of virtual network embedding algorithm needs to be intensive studied.By analyzing the 400 GB dataset published by Google cloud datacenters in November 2014 for more than 100 million network service requests,this paper refines the characteristics of network requests,and provides a model of evaluating virtual network embedding algorithms for network service requests in real network environments.In view of the previous research,there is no comparison between the actual performances of the common virtual network embedding algorithm.This paper designs the experiments to compare the performance of two virtual network embedding algorithms based on integer linear programming in real network environments with different scales,the purpose is to analyze the computing performance of each algorithm in real network environment,and to make some guidance to the choice of virtual network embedding algorithm in practical applications.Besides,the prediction model of virtual network requests establishing in the experiments has also played a very important role in the future research.In addition,according to the observation of Google cloud dataset in this paper,a large number of requests are arrived in batch form in practical applications,while the traditional virtual network embedding algorithm has some limitations in processing the requests of batch arrival.Therefore,this paper proposes a heuristic algorithm based on genetic algorithm called VNE-NGA algorithm,which focuses on solving the virtual network request arriving in batch form.Through theoretical analysis and experiments,it is concluded that the receiving rate of the virtual request and the long-term income generated by the underlying network are much higher comparing the VNE-NGA algorithm to traditional algorithm in the real network environment.