SA-DDQN: Self-Attention Mechanism Based DDQN for SFC Deployment in NFV/MEC-Enabled Networks

Network function virtualization (NFV) is able to reduce the delay and improve the flexibility of network services in mobile edge computing (MEC) networks via deploying the service function chain (SFC) that consists of a sequence of ordered virtual network functions. However, it is still challenging to deploy SFC with delay guarantees and resource efficiency while taking into account the real-time network variations and dispersed edge server nodes in NFV/MEC-enabled networks. To address the issue, this paper proposes a self-attention mechanism-based double deep Q-network algorithm (SA-DDQN) for SFC deployment to jointly minimize the resource consumption on servers and bandwidth consumption on links within delay limits in dynamic NFV/MEC-enabled networks. In particular, we introduce the self-attention mechanism in the deep neural network structure, which enables the agent to pay its attention on more valuable physical nodes when making decisions, thus improving the efficiency of SFC deployment. Additionally, we utilize the Markov decision process (MDP) model to solve the problem of real-time network state variations. Finally, extensive simulation results show that our proposed SA-DDQN SFC deployment algorithm can reduce resource consumption by 25% and delay by 18.4% compared with the state-of-the-art algorithm.