Markdown Version | Session Recording

Session Date/Time: 20 Jan 2026 08:00

NMRG

Summary

The NMRG session focused on presenting and discussing extensions to two existing frameworks: I2NSF (Interface to Network Security Functions) and I2ICF (Interface to In-Network Computing Functions). The primary goal was to inform NMRG participants about the roadmaps and ongoing work for these frameworks and to explore the potential role of NMRG in their future development, particularly from a research perspective. Discussions highlighted the need for clear articulation of research questions, differentiation from existing work, and showcasing implementation efforts.

Key Discussion Points

The meeting was structured into two main parts, followed by open discussion. Co-chairs Jérôme François and Jefferson welcomed participants and outlined the IRTF notewell and IPR disclosure rules.

I2NSF (Interface to Network Security Functions)

Paul (Jung-Soo Park) provided an overview of I2NSF and its proposed extensions.

• I2NSF Document Status and Roadmap:

  • Six I2NSF documents (applicability, capability, and four interface Yang data models) are currently in the RFC-Editor Queue, with editorial issues being resolved. Expected completion by February.
  • Motivation for augmentation is to align I2NSF with the Intent-Based Networking (IBM) concept defined in RFC 9315.
  • Proposed roadmap: officially present three new drafts in NMRG, decide on NMRG adoption, and publish as informational RFCs.

• I2NSF for Security Management Automation:

  • Patrick presented a draft proposing extensions to I2NSF for IBM, focusing on closed-loop security control. This includes an analytics interface for policy reconfiguration and feedback, and a security audit system for detecting insider/supply chain attacks via remote attestation.
  • Discussion: A participant inquired about implementation status. Paul confirmed that basic I2NSF and closed-loop security control concepts had been implemented as Proof-of-Concepts (POCs) in IETF Hackathons, with source code available on GitHub. However, the security audit system is currently at a design level. It was suggested that future presentations should highlight scientific research aspects, such as algorithms used and novelty compared to existing closed-loop security frameworks.

• I2NSF Analytics Interface Yang Data Model:

  • Patrick presented a draft detailing the Yang data models for the analytics interface. This interface facilitates policy reconfiguration (analyzer suggests changes to security controller) and feedback information (analyzer provides user reports, e.g., for resource issues).
  • Discussion: The co-chair reiterated that NMRG is a research group, not a standardization body. The focus should be on research aspects like the algorithms for identifying security issues and generating policies, rather than solely on interface definitions and Yang models, which are more suited for IETF working groups (e.g., network management or security WGs). The importance of highlighting novelty in the context of existing closed-loop security controls was emphasized.

• Integrated Security Service System for 5G Networks using I2NSF:

  • Joseph presented a draft on integrating I2NSF into 5G core networks, particularly for intent-based security management, with a focus on edge-based security. The proposed architecture embeds I2NSF components into 5G service architectures, placing security functions closer to the UPF to reduce latency compared to a centralized cloud approach. The work also covers policy migration during UE handovers.
  • Discussion: A participant asked about testbed results, particularly concerning latency reduction. Joseph clarified that the testbed implementation is in progress, and results, including latency checks, are expected within the year. It was suggested to include information about implementation status and pointers (e.g., GitHub links) in presentations to encourage community contribution.

I2ICF (Interface to In-Network Computing Functions)

Paul (Jung-Soo Park) introduced I2ICF and its proposed extensions.

• I2ICF Problem Statement:

  • I2ICF aims to create a standard framework and interfaces for managing AI agents and in-network computing functions (ICFs) in IoT environments, including on-device AI and moving objects (e.g., robots, software-defined vehicles). This work aligns with IBM.
  • Proposed roadmap: suggest NMRG adoption, revise drafts, and publish as informational RFCs.
  • Discussion: A co-author (Kehan) highlighted the difference from I2NSF, focusing on management functions and agent-to-agent (A2A) protocols, noting Google's A2A work. Another co-author (Khan) expressed strong support for I2ICF in NMRG, citing its relevance to network management, emerging use cases (e.g., in 3GPP offloading compute to terminals/UPF), and the growing interest in agentic AI.
  • Co-chair Feedback: The co-chair emphasized the need to attract support from group members, clearly differentiate research questions from standardization aspects, and provide a review of the state-of-the-art in distributed computation and multi-agent frameworks to highlight the novelty of the I2ICF proposals.

• I2ICF Framework:

  • Byung Manan presented a draft proposing a framework for managing and configuring ICFs, essential for NFV and SDN environments. The framework applies to edge cloud environments (e.g., P4 switches, VNF-based consensus) and mobile objects (on-device AI, vehicles), integrating wired and wireless networks with consistent management under mobility. It uses a closed-loop system for intent translation and monitoring.
  • No specific questions were recorded for this presentation in the transcript.

• Cooperative Intelligent Transportation Systems (C-ITS) using I2ICF:

  • Byung Manan presented a use case for I2ICF in C-ITS, focusing on how agent AI can support real-time coordination and decision-making in environments with multiple interacting entities (vehicles, pedestrians, RSUs). The draft highlights the need for intent-based management in C-ITS due to increasing traffic density and complexity, enabling dynamic selection of communication paths and policies. Real-world deployment screenshots were shown.
  • No specific questions were recorded for this presentation in the transcript.

• Intent Translation Framework for IoT Networks:

  • Mozi Gu presented a draft on an intent translation framework for IoT networks. The motivation is to bridge the semantic gap between natural language human intents and precise, machine-executable network policies. The proposed architecture uses an intent extractor, semantic mapper (knowledge graph embedding), intent resolver (reasoning), and policy composer, aiming to avoid "hallucination" common in large language models.
  • Discussion: The co-chair commented on the draft's focus on IoT, suggesting the framework appears generic enough to apply beyond IoT. It was also noted that the architecture resembles Retrieval Augmented Generation (RAG) systems. A global comment was made about the general lack of publicly available datasets and benchmarks for intent translation research across the IETF community. The author clarified the IoT focus was to narrow down the use case and fit within the I2ICF framework's AI agent focus.

Collective Guidance and Future Work

No formal decisions were made to adopt any of the drafts. The co-chairs provided collective guidance to all presenters and the broader group:

• Mailing List Engagement: Authors are strongly encouraged to use the NMRG mailing list to discuss their drafts, gather feedback, and attract support from other members before the next IETF meeting.
• Research Focus: Clearly articulate the specific research questions addressed by the work. Differentiate the proposed research contributions from existing state-of-the-art solutions in relevant fields (e.g., distributed computing, multi-agent systems, closed-loop security, intent translation).
• Implementation & Evaluation: Where applicable, present implementation results, prototypes, demos, or preliminary testbed evaluations (e.g., from IETF Hackathons). Providing pointers to codebases (e.g., GitHub) is highly encouraged to facilitate community involvement.
• Standards vs. Research: Clearly distinguish which aspects of the work are research-oriented (suitable for IRTF) and which might lead to standardization efforts (more appropriate for relevant IETF Working Groups, e.g., Network Management or Security WGs, or other IRTF Research Groups like T2TRG for IoT).
• Presentation Style: For future NMRG sessions, if a cluster of drafts is proposed, consider aggregating individual presentations into a more concise, overarching presentation to better highlight the collective research value and allow more time for discussion.

The chairs acknowledged the value of the presented topics for research and expressed their commitment to facilitate discussions and provide meeting time in the future, subject to community interest and alignment with NMRG's research mandate.