Graph neural networks for predicting network characteristics in New IP and ManyNets architectures

In New IP and ManyNets architectures (ITU-T Network 2030), the need to predict network characteristics, including path delay, without heavy simulation grows; it remains unclear when graph neural networks outperform simple computational methods and how such models generalize to different graph sizes. This article aims to assess applicability of a graph neural network to the path delay task on synthetic graphs with a formula accounting for link load, and to evaluate generalization to larger graphs. A comparative experiment on Erdős–Rényi graphs was applied: a graph convolution-based model was compared with a baseline method; two experiments were conducted: a load-aware target latency experiment and a test on graphs with 15 and 20 nodes after training on graphs with 15 nodes. Results (single run): in the first experiment the baseline gave MAE 1.85 and MAPE 7.89 %, the graph model 9.91 and 59.20 %; in the second, when moving from 15- to 20-node test graphs, the graph model’s MAE decreased by about 7 % and the baseline’s increased by about 8 %. The approach is concluded applicable on synthetic data as a first step toward models for predicting network characteristics in New IP and ManyNets architectures. The materials are of practical value for specialists when choosing and validating delay prediction methods and planning experiments on synthetic topologies.

1. Rusek K., Suárez-Varela J., Almasan P., Barlet-Ros P., Cabellos-Aparicio A. RouteNet: Leveraging graph neural networks for network modeling and optimization in SDN. IEEE Journal on Selected Areas in Communications. 2020;38(10):2260–2270. https://doi.org/10.1109/JSAC.2020.3000405

2. Kipf Th.N., Welling M. Semi-supervised classification with graph convolutional networks. In: 5th International Conference on Learning Representations, ICLR 2017, 24–26 April 2017, Toulon, France. 2017. URL: https://openreview.net/forum?id=SJU4ayYgl

3. Scarselli F., Gori M., Tsoi A.Ch., Hagenbuchner M., Monfardini G. The graph neural network model. IEEE Transactions on Neural Networks. 2009;20(1):61–80. https://doi.org/10.1109/TNN.2008.2005605

4. Gilmer J., Schoenholz S.S., Riley P.F., Vinyals O., Dahl G.E. Neural message passing for quantum chemistry. In: Proceedings of the 34th International Conference on Machine Learning, ICML 2017, 06–11 August 2017, Sydney, NSW, Australia. PMLR; 2017. P. 1263–1272.

5. Xu K., Hu W., Leskovec J., Jegelka S. How powerful are graph neural networks? In: International Conference on Learning Representations, ICLR 2019, 06–09 May 2019, New Orleans, LA, USA. 2019. URL: https://openreview.net/forum?id=ryGs6iA5Km

6. Farreras M., Soto P., Camelo M., Fàbrega L., Vilà P. Improving network delay predictions using GNNs. Journal of Network and Systems Management. 2023;31(4). https://doi.org/10.1007/s10922-023-09758-9

7. Roy A., Pachuau J.L., Saha A.K. An overview of queuing delay and various delay based algorithms in networks. Computing. 2021;103(10):2361–2399. https://doi.org/10.1007/s00607-021-00973-3

8. Knight S., Nguyen H.X., Falkner N., Bowden Rh., Roughan M. The internet topology zoo. IEEE Journal on Selected Areas in Communications. 2011;29(9):1765–1775. https://doi.org/10.1109/JSAC.2011.111002

9. Erdos P., Renyi A. On the evolution of random graphs. Publications of the Mathematical Institute of the Hungarian Academy of Sciences. 1960;5:17–61.

10. Fey M., Lenssen J.E. Fast graph representation learning with PyTorch Geometric. arXiv. URL: https://arxiv.org/abs/1903.02428 [Accessed 5th March 2025].