DDQN Based Slot Allocation for Flexible Ethernet

• Published in: 2022 6th Asian Conference on Artificial Intelligence Technology (ACAIT) pp. 1 - 5
• Main Authors: Zhang, Hao, Zhang, Kun, Li, Yan, Fu, Xiaoying, Gu, Rentao, Miao, Jiansong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 09.12.2022
• Subjects: double deep Q-network; Ethernet; Flexible Ethernet; Heuristic algorithms; Markov decision process; Network slicing; Packet loss; Reinforcement learning; Simulation; slot allocation; Transportation
• DOI: 10.1109/ACAIT56212.2022.10137913

F1exible Ethernet (FlexE) is a new technique in transportation network which can provide more flexible resource management by network slicing and service isolation. With the increasing of service types, it is a challenging work to design a high efficiency multi-service resource allocation scheme to meet users' different demands, such as latency, packet loss rate, etc. In this paper, we propose a new slot allocation scheme based on deep reinforcement learning (DRL) method to achieve FlexE multi-service oriented resource allocation by optimizing the queuing time and packet loss rate. This optimization problem is formulated as a Markov decision process (MDP) and it can be solved by double deep Q-network(DDQN) algorithm. Numerical results demonstrate that the proposed scheme outperforms the benchmark scheme in convergency, latency and packet loss rate.