Parallel Deployment of Service Function Chains Based on Network State Prediction

Network Function Virtualization (NFV) transforms network functions into virtualized instances to enhance the flexibility, reliability, and scalability of the network, reduce network deployment and maintenance costs, and improve service quality and flexibility. Service Function Chains (SFC) has also become a popular form of network services with the development of NFV, allowing network traffic to pass through a series of virtualized network functions in a specific order. The deployment of SFC has become a research hotspot in NFV. Because the deployment of SFC requests depends on the current network state and the network state changes after deployment, there is a certain topological dependency among the deployments of multiple requests. Therefore, many recent research works have adopted a serial approach to deploy multiple requests one after another, which requires more response time to handle burst traffic. This paper proposes a parallel deployment algorithm based on the Seq2Seq model for network state prediction. This way, the deployment of each request only depends on the predicted network state by the model, breaking the topological dependency among deployments of multiple requests, and enabling the simultaneous deployment of multiple requests. We trained the Seq2Seq prediction model on networks of various scales and modified the existing serial algorithm to a state prediction-based parallel algorithm. Experimental results demonstrate that compared to the serial algorithm, the proposed algorithm reduces the average response time for deploying burst traffic by 2.52-3.94 times, while also exhibiting good robustness in physical networks of different scales.