Markdown Version | Transcript | Session Recording | Session Materials

HOTRFC

Summary

The HOTRFC (Hot Request for Comments) session at IETF 126 featured a full agenda of 21 lightning presentations. The presentations covered a wide range of emerging proposals, research, and technical challenges across physical mapping, BGP security, AI network optimization, DNS developments, L4S congestion control, IoT deployment, identity delegation, and confidential computing.

The session was designed as a rapid-fire forum to encourage cross-community collaboration, solicit feedback, and direct new work to relevant IETF working groups, research groups, or side meetings. No formal consensus was determined during the session.

Key Discussion Points

1. Physical Layer and Fiber Mapping

• 01 Open Fibre Data Standard (OFDS)
  • Presenter: Dan York
  • Technical Details: York highlighted the lack of standardized, open physical-layer maps for land-based fiber infrastructure (often locked in PDFs, JPEGs, or proprietary KML/Excel sheets). The Open Fibre Data Standard (OFDS) defines Points of Presence (PoPs) and network spans using JSON and geopackages. The goal is to bring this specification into the formal SDO standardization process at the IETF (drawing parallels to GeoJSON and GeoPriv work).
  • Side Meeting Info: A side meeting was scheduled for Wednesday at 2:45 PM in the room adjacent to Grand Klimt Hall.

2. Routing Security and Path Validation

• 02 BGP Secure Origin Delegation Authorization (SODA)
  • Presenter: Shengmen Jing (Forest)
  • Technical Details: Presenting on BGP SODA, Jing addressed the issue of false positives in Route Origin Authorizations (ROAs) when multiple legitimate ASes announce the same prefix (e.g., in commercial partnerships). SODA introduces a new BGP path attribute. The delegating AS signs a "sudo token" using its router key and attaches it to BGP updates, enabling receivers to explicitly verify origin delegation.
• 03 PAVA
  • Presenter: Maic Sans Fleishier
  • Technical Details: Fleishier introduced Path Validation (PAVA) as a complementary mechanism to ASPA to prevent route leaks and hijacks. Under PAVA, the received AS path is broken into segments of three ASes. The validating AS queries the involved ASes via DNS to check segment validity. This allows ASes to maintain validation information locally in their own DNS.

3. AI and Data Center Network Efficiency

• 04 Beyond Throughput
  • Presenter: Nengye Ye
  • Technical Details: Ye introduced a "switching efficiency" ($\eta$) framework to measure network efficiency during LLM training. Traditional throughput/bandwidth metrics show activity but not whether the traffic translates to effective computation (e.g., intermediate data discarded in ring-based reductions vs. in-network computing). The framework divides computationally effective data retained per unit of time by the total provisioned switching capacity, decomposing efficiency into Port Utilization, Routing Efficiency, and Data Efficiency.
• 21 From Fragmented Benchmarks to an Interoperable Format for Evaluating Network LLM Agents
  • Presenter: Zicheng Wang (Cheng Wang Gu)
  • Technical Details: Wang detailed the lack of interoperability among testing benchmarks for network LLM agents. The proposed benchmark framework uses an extensible format composed of datasets, environments, agents, and evaluators. Tasks are represented as five-part schemas (intent, topology, initial config, ground-truth solutions, and test cases), with an experimental harness showing significant reductions in execution time and token usage.

4. DNS-Based and Cryptographic Specifications

• 05 DIVE: A DNS-based Alternative to CA-Based Code Signing for HTTP Artifacts
  • Presenter: Mateo Karék (Remote)
  • Technical Details: Karék proposed DIVE, which stores public keys in DNSSEC-protected DNS records rather than relying on complex, costly CA-based code-signing certificates. Origin servers attach signatures and key identifiers directly in HTTP headers; clients look up public keys via DNSSEC to verify software/firmware updates. It supports key rotation, subdomain restrictions, and a "report-only" migration mode.
• 06 Self-Certifying Identity and Capability-Based Delegation for Autonomous AI Agents
  • Presenter: Andrzej Duda
  • Technical Details: Duda presented a framework that assigns self-certifying identities (hashes of public keys representable as DNS names) to AI agents. Binds between identities and public keys are maintained in a mutable store (DNS/DNSSEC or IPFS/IPNS DHT), allowing secure capability-based delegation from users to agents executing network security tasks.
  • Chat Commentary: Stuart Card noted that this approach resembles Hierarchical Host Identity Tags (HHITs) but is tied primarily to DNS rather than IPv6 addresses, extended to handle capabilities.
• 07 PQC DNSSEC MTL Updates
  • Presenter: Andrew Fregly
  • Technical Details: Fregly provided an update on migrating DNSSEC to Post-Quantum Cryptography (PQC) using the Merkle Tree Ladder (MTL) approach to address the overhead of large PQC keys/signatures (such as ML-DSA and SLH-DSA). Since IETF 124, the MTL specification has been made self-contained and updated to use ML-DSA, introducing EDNS(0) signaling to negotiate when to return full vs. condensed signatures.

5. Congestion Control and Transport

• 08 Interactions between L4S and FQ-based Queue Disciplines
  • Presenter: Mohit
  • Technical Details: Mohit discussed the friction of co-deploying Flow Queuing (FQ-Codel, CAKE, FQ-PI) and L4S. FQ focuses on per-flow isolation for fairness, while L4S leverages a coupled dual-queue structure to achieve low latency. How these mechanisms interact—such as the impact of FQ's Deficit Round Robin scheduler on L4S congestion marking and sender pacing—is not yet understood and requires fine-grained experimental data.

6. AI Inferences and Application Protocols

• 09 AI Inference Streaming: The Protocol Gap
  • Presenter: Santosh
  • Technical Details: Santosh highlighted a critical standardization gap in streaming LLM tokens. While the transport layer (HTTP/2, TLS, SSE) is standardized, the payloads above them use highly fragmented schemas (e.g., distinct Server-Sent Events schemas by OpenAI/Google, WebSockets, or Connect/Protobuf over HTTP). This forces middleware and compliance auditors to build custom parsers for each vendor. The presenter proposed a standard Type-Length-Value (TLV) JSON/binary schema over SSE.
• 14 HTTP Events Query
  • Presenter: Rahul (Remote)
  • Technical Details: Rahul proposed "Events Query," utilizing the newly standardized HTTP QUERY method to enable structured, real-time push notifications. Current alternatives, like Server-Sent Events (SSE), do not support custom headers, are unmaintained, or push complexity to WebSockets. Events Query allows clients to negotiate media types for both notifications and representation, facilitating clean stream termination and intermediary caching.

7. Cyber Security & AI Resolver Impacts

• 10 Practical Cyber Security
  • Presenter: Michael P.
  • Technical Details: Representing the UK National Cyber Security Centre, the presenter advertised the "Practical Cyber Security" side meeting, aimed at exchanging practical insights on changing global cyber threats and defending heterogeneous environments.
  • Chat Commentary: Chris Box noted that the side meeting was scheduled for Wednesday at 4:00 PM in Grand Klimt 3.
• 11 AIDNS: Measuring Security and Stability Impacts of AI on the DNS Root
  • Presenter: James Kunle Olurandari
  • Technical Details: Olurandari highlighted the need to study how the proliferation of autonomous "agentic" AIs impacts DNS resolvers. He called for collaborative empirical studies to gather data on AI-generated resolver traffic patterns, preventing potential stability issues at the DNS root and resolver layers.

8. Agentic Commerce, Identity, and Consent

• 12 KYAPay: Verified Agent Identity and Payments for Practical Interoperable Agentic Commerce on the Existing Web
  • Presenter: Mike Jones
  • Technical Details: Jones pitched "KYAPay" to enable autonomous agents to transact on the existing web without being blocked by anti-bot/CAPTCHA checks. It utilizes a KYAPay token—formatted as a JSON Web Token (JWT) passed via an HTTP header—to securely represent the authorizing user, the executing agent, and payment claims.
• 13 Agent Accountability, Composed
  • Presenter: Steven Meyer
  • Technical Details: Meyer proposed an "action digest" approach to compose multiple agent verification tools. By using standardized "agent action capsules," verifications of identity, authorization, and receipts can be ledgered into SCITT/RATS architectures, creating transparent, auditable history trails to establish agent reputation.
• 18 Who Authorizes the Request?
  • Presenter: Yuning Zhang
  • Technical Details: Zhang discussed the security risk of malicious intermediate agents modifying user requests (e.g., changing a $50 refund to $5,000) when standard access tokens only validate the agent's identity. He proposed an "Intent Admission Point" to generate signed Intent Admission Assertions (IAAs) verifying user consent and origin details, which the agent must deliver to the execution endpoint.

9. Confidential Computing

• 15 Confidential Computing and Digital Sovereignty
  • Presenter: Muhammad Usama Sardar
  • Technical Details: Sardar presented evidence of structural vulnerabilities in cloud attestation and trust mechanisms (citing recent high-severity and critical CVE disclosures). Due to hardware supply chains and relay vulnerabilities, attestation lacks true accountability. He advocated for a dedicated IRTF research group to study the architectural flaws of digital sovereignty in confidential computing.
  • Chat Commentary: Lixia Zhang and Muhammad Usama Sardar discussed utilizing the Decentralized Internet Infrastructure Research Group (DINRG) mailing list as a potential starting venue to coordinate research on confidential computing.

10. Resilient Control Planes and Network Management

• 16 KIRA – Routing for Resilient Control Planes
  • Presenter: Roland
  • Technical Details: Roland addressed major outages caused by control plane misconfigurations (e.g., the Facebook 2021 outage). KIRA is a scalable (supporting hundreds of thousands of nodes), zero-touch, self-configuring routing architecture designed specifically for control planes. It includes distributed key-value stores for rendezvous services and runs over IPv6 without configuration.
• 17 Problem Statement and Roadmap for BGP FlowSpec Monitoring
  • Presenter: Yujia Gao
  • Technical Details: Gao detailed the monitoring gap in BGP FlowSpec, where route propagation succeeds but enforcement fails in the data plane (due to policy conflicts or installation errors). The proposed monitoring framework spans the intent, control, and data planes using extensions to BMP, IPFIX, and YANG.
• 19 Framework for Normalizing Multi-Vendor Network Inputs for LLM-Assisted Network Management
  • Presenter: Shailesh
  • Technical Details: Shailesh proposed an input normalization layer between network devices and LLMs. Since multi-vendor devices represent the same operational intents differently (e.g., "show interface" vs. "display interfaces"), a normalizing classifier maps inputs into standardized schemas to generate consistent, predictable prompts for LLM parsers.
• 20 Application-Centric Deployment across the IoT Computing Continuum
  • Presenter: Carlos
  • Technical Details: Carlos addressed the challenge of deploying and adapting service logic across the edge-to-cloud IoT computing continuum (e.g., drone collision avoidance DAGs). He proposed a choreography-based deployment draft where service graphs are distributed directly to nodes, allowing self-organization and service migration, potentially assisted by AI agents.

Decisions and Action Items

• No formal decisions were made during this session. HOTRFC serves purely as a venue for soliciting feedback, pitching early proposals, and advertising side meetings.

Next Steps

• OFDS: Interested parties are encouraged to contact Dan York (york@isoc.org) or Steve Song (song@isoc.org).
• PQC DNSSEC MTL: Developers and researchers are invited to review the open-source code and MTL draft, or connect at the Hackathon demo hour.
• L4S & FQ Coexistence: Mohit requested network operators and researchers with simulation or deployment tools to collaborate on fine-grained scheduling experiments.
• AI Inference Streaming: Santosh sought co-authors to develop an IETF draft proposing a standardized TLV JSON/binary payload over SSE.
• Confidential Computing / DINRG: Muhammad Usama Sardar will coordinate with Lixia Zhang and Dirk Kutscher on the DINRG mailing list to discuss digital sovereignty and confidential computing research.
• KIRA: Roland requested operators and researchers to evaluate the KIRA zero-touch control-plane routing draft in the RTGWG.
• BGP FlowSpec Monitoring: Yujia Gao scheduled a detailed presentation on Friday in the IDR working group to collect feedback.
• LLM Multi-Vendor Normalization: Shailesh invited contributors to review the normalization framework draft.