Research On Virtual Network Embedding Model And Algorithm Based On Graph Neural Network

At present,5G networks are widely used and 6G networks are receiving much attention in research,but the development of hardware networks is slow and network virtualization technologies need to be studied to achieve resource sharing,where the virtual network embedding problem is a key challenge.In order to meet and balance the needs of network service providers and users,the designed virtual network embedding algorithms need to maximize the revenue of network service providers and the experience of users.The optimization objectives for existing schemes usually focus on a single metric or a simple combination of multiple metrics,which makes it difficult to reasonably embed virtual network requests,and the experimental results are not satisfactory.In this paper,two virtual network embedding algorithms based on graph neural networks are proposed:Firstly,in this paper,we propose VGAE-VNE,a virtual network embedding algorithm based on the variogram autoencoder.simple embedding representations can be learned by the variogram autoencoder.These representations aggregate the characteristics of the nodes themselves as well as the characteristics of the network structure,and then we use K-means clustering method to distinguish the obtained embedding representations.The simulation experiments demonstrate that the VGAEVNE algorithm performs well under the evaluation metrics of acceptance rate,longterm average revenue,and long-term average CPU resource utilization,and outperforms other comparative algorithms in terms of comprehensive performance.Secondly,this paper proposes a virtual network embedding algorithm GAT-VNE based on the graph attention mechanism,where the scale of the physical and virtual networks faced is often relatively large in real situations.Although the representation of nodes can be obtained by encoding through graph convolutional neural network,it ignores the relationship between nodes and nodes when encoding,which leads to the embedding result may not reach the optimum.Thus,this paper proposes a graph attention mechanism-based virtual network embedding algorithm GAT-VNE,which measures the relationship between nodes by introducing an attention mechanism to determine the weights between nodes.And in the node embedding stage,a node resource capacity quantification method is proposed to quantify the node embedding potential.Not only the resource capacity of nodes but also the resource capacity of the whole network is considered.Through simulation experiments,we verify the performance of the GAT-VNE algorithm at different network sizes,which has a reasonable bandwidth and node allocation method,thus making it able to support the embedding of virtual network requests in a stable way in the long term.