A Graph Attention-Enhanced Deep Reinforcement Learning Framework for Dynamic Service Function Chains Orchestration
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 24386-24397
DOI: 10.1109/tmc.2026.3734450 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Chuangchuang Zhang und 9 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Fundamental Research Funds for the Central Universities, JSPS KAKENHI.
Abstract (Originalsprache)
In NFV-enabled future networks, dynamic Service Function Chain (SFC) orchestration remains challenging, as existing methods suffer from low deployment success rate under high service load, insufficient utilization of inherent graph structure of network topology, poor handling of sequential Virtual Network Function (VNF) dependencies, and unstable performance in highly dynamic environments. These limitations severely restrict the overall resource efficiency and long-term revenue of network operators. To address this gap, we propose GADS, a Graph Attention-enhanced Deep Reinforcement learning (DRL) framework for dynamic SFC orchestration to improve deployment success rate and resource utilization. Specifically, a Graph Convolutional Network (GCN)-based SFC splitting mechanism is designed to automatically split the SFC by capturing real-time network topology status and implicit SFC demand representations. Then, a graph attention-enhanced DRL-based SFC deployment mechanism composed of a graph attention network driven placement algorithm and a GCN and pointer network-based routing algorithm, is designed to make joint optimal decisions for VNF placement and SFC routing. Simulation results demonstrate that the proposed mechanism outperforms benchmark mechanisms in terms of deployment success rate, revenue-cost ratio, and resource utilization.
Förderung
- National Natural Science Foundation of China
- Fundamental Research Funds for the Central Universities
- JSPS KAKENHI
Zitieren
APA
Zhang, C., Li, H., He, Q., Li, F., Wang, X., Zheng, K., Hawbani, A., Yu, K., Lv, J., & Xu, L. (2026). A Graph Attention-Enhanced Deep Reinforcement Learning Framework for Dynamic Service Function Chains Orchestration. IEEE Transactions on Mobile Computing, 25(12), 24386-24397. https://doi.org/10.1109/tmc.2026.3734450
Vancouver
Zhang C, Li H, He Q, Li F, Wang X, Zheng K, et al. A Graph Attention-Enhanced Deep Reinforcement Learning Framework for Dynamic Service Function Chains Orchestration. IEEE Transactions on Mobile Computing. 2026;25(12):24386-24397. doi:10.1109/tmc.2026.3734450
BibTeX
@article{Zhang2026,
author = {Zhang, Chuangchuang and Li, Haodong and He, Qiang and Li, Fuliang and Wang, Xingwei and Zheng, Kaifa and Hawbani, Ammar and Yu, Keping and Lv, Jianhui and Xu, Li},
title = {A Graph Attention-Enhanced Deep Reinforcement Learning Framework for Dynamic Service Function Chains Orchestration},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {24386--24397},
doi = {10.1109/tmc.2026.3734450}
}
EUVIMED