16th SIG-NGN Meeting, in collaboration with SIG-AI
SIG-NGN is the Special Interest Group on Next Generation Networks.
The 16th SIG-NGN meeting, in collaboration with SIG-AI, will take place on the 9th and 10th October 2025, co-located with the LHCOPN and LHCONE meeting at KIT in Karlsruhe, Germany. This will be an in-person meeting over two half-days with remote participation.
Theme: Networking for AI and AI for Networking.
One half-day will be spent exploring how we build networks for new upcoming demands, such as ‘AI,’ and we hope to include presentations on Ultra Ethernet, beyond 1T (or even 10T) networking, and “ultra large networks.”
The second half day will be spent looking at how AI (or machine learning) can help us build and operate the networks. In this we hope to have a couple of (suitably prepared) vendor presentations, plus community presentations on how the technology might enable troubleshooting and workflow management.
<<the link for the recordings will be available in the SIG-NGN wiki page>>
Photos
Agenda - (all times are in CEST - UTC+2)
Thursday, 9 October 2025
| 12:45 - 13:00 | Get a coffee - Testing audio & video | |
|---|---|---|
| 13:00 - 13:10 ('10) | Welcome and Introduction Rob Evans, Jisc | |
| 13:10 - 13:20 ('10) | An overview of SIG-AI | |
| 1st session | ||
13:20 - 13:40 ('20) | Why do we need to talk about networking for AI? Requirements and solutions. Kamran Naqvi, Broadcom | |
13:40 - 14:00 ('20) | AI at Scale: HPC Primitives in the Cloud-Native Era Ricardo Rocha, CERN | |
14:00- 14:30 ('30) | Building Networks for AI Eli Dart, ESnet | |
| 14:30 - 15:00 ('30) | Coffee Break |
|
| 2nd session | ||
| 15:00 -15:30 ('30) | UltraEthernet Andreas Roeder, Nokia | |
| 15:30 - 16:00 ('30) | Tour into Practical AI for Network Traffic Analysis Jaroslav Pesek, CESNET We present recent R&D results within an end-to-end monitoring and analysis pipeline. We show how to detect network anomalies and convert them into events for SOC team, introduce libraries and datasets for reproducible modelling, and demonstrate that a simple k-NN on packet-sequence metadata can perform on-par with complex models. We conclude with visualisations that aid traffic interpretation. | |
| 16:00 - 16:30 ('30) | AI for Network Management and the IETF Network Management Research Group Jefferson Nobre | |
| 16:30 - 17:00 ('30) | Discussions |
|
End of the day |
|
Friday, 10 October 2025
| 8:45 - 9:00 | Get a coffee - Testing audio & video | |
|---|---|---|
| 3rd session | ||
9:00 - 9:10 ('10) | Scene setting: AI for network management Lars Fisher, NORDUnet |
|
9:10 - 9:40 ('30) | GN5-2 Survey on AI for Network Management Pavle Vuletic |
9:40 - 10:10 ('30) | AI for Workflow Orchestrator Peter Boers, SURF | |
|---|---|---|
10:10 - 10:30 ('20) | LogLLM Dimitris Pantazatos, NTUA | |
| 10:30 - 11:00 ('30) | Coffee Break |
|
| 4th session | ||
| 11:00 -11:20 ('20) | Classification and Identification of network traffic using AI models Maarten Meijer | |
| 11:20 - 11:40 ('20) | AI/ML data center fabric advanced load balancing Michal Styszynski, Juniper | |
| 11:40 - 12:10 ('30) | Discussions |
|
| 12:10 - 12:30 ('20) | Wrap-Up |
|
End of the meeting |
|