Research And Implementation On Routing Mechanism Of Cross-domain Computing Network Based On GNN

Computing network is a new type of information infrastructure that integrates cloud,edge,terminal and other computing resources.It is also able to schedule these computing resources uniformly according to requirements.A large-scale computing network may contain multiple autonomous systems.There may be several autonomous systems controlled by different internet service providers.Also,there may be several autonomous systems controlled by one internet service provider but can’t communicate with each other.In such a scenario,to meet the requirements of the domain,computing devices in other domains may need to be deployed,which means,cross-domain scheduling is required.However,computing demands include bandwidth on demand and computing power on demand,and the topology and network states in each domain are different.What’s more,the information in each domain needs to be kept secret from other domains.Therefore,cross-domain scheduling computing equipment is complicated.If simply using protocols to schedule,taking multidimensional requirements into account will be not easy.But the existing machine learning models have poor generalization,that is,they have poor performance when dealing with topologies not seen in training.So,they cannot be applied to domains with different topologies.To solve the problems mentioned above,this paper proposes a machine leaning model worked on multi-domain computing network based on Graph Neural Networks(GNN)and Deep Q Network(DQN).Because of the excellent generalization of GNN,this model can well deal with the topology unseen in the training process,and generate multiple alternative scheduling paths based on the computational demand and the topology.Then the model uses the decision-making ability of DQN to evaluate these paths,so as to select the most appropriate scheduling path.In addition,a multi-domain computing network based on Mininet and RYU is built to test the model performance.As a centralized architecture,the network has three layers:network environment,intra-domain controller layer,and inter-domain controller layer.The network environment contains multiple domains.Each domain is controlled by an intra-domain controller,and all these intra-domain controllers are controlled by the inter-domain controller.Each controller uses the GNN+DQN model for scheduling.In the end,the experimental results show that,compared with the simple DQN model and Dijkstra algorithm,the proposed model has a shorter task completion time and a more balanced network load when scheduling.