Adaptive Network Function Chain Orchestration Strategy for 6G Inclusive Intelligent Services

Abstract

In the upcoming 6G era, inclusive intelligent services(IISs) that rely on integrated communications and AI arithmetic will become the norm. These services require efficient distributed intelligent learning or reasoning. However, with the proliferation of complex applications, providing differentiated and customized services through effective network function chain (NFC) orchestration has become a significant challenge. In this paper, an adaptive NFC orchestration strategy (ANFCOS) is proposed for 6G IIS to minimize costs. In particular, a tailored NFC orchestration is optimized, catering to diverse quality-of-service requirements. ANFCOS models this process as a Markov decision process and a multiactor-attention-critic (MAAC) reinforcement learning method for solving. The strategy enables the deployment and optimization of operational costs of multiple parallel NFCs, relying on a dynamic mechanism of shifting attention. Additionally, We introduced a reinforcement learning framework based on maximum entropy to enhance the MAAC for solving and demonstrate its convergence. The simulation results show that ANFCOS significantly improves the network performance of 6G intelligent services. Compared with MADDPG, DDPG and MCTS, the cost of orchestration was reduced by 8.2%, 22.1% and 28.6%; the total delay of orchestration was reduced by 7.7%, 12.7% and 10.8%