Session Date/Time: 15 Jul 2022 15:00

RAW

Summary

The RAW Working Group held an interim meeting to discuss the architecture draft and follow up on previous action items. The primary technical discussion revolved around the interaction between RAW/DetNet and upper-layer transport protocols. A key concern was the inadequacy of current general-purpose transport protocols (like TCP or UDP) to fully leverage the deterministic paths provided by DetNet and RAW. Participants discussed the need for a "deterministic transport protocol" or significant adaptations to existing ones. Decisions were made to update the RAW architecture document with explicit dependencies on transport behavior and to initiate broader discussion in the DetNet Working Group on this transport layer gap.

Key Discussion Points

RAW Architecture Document Updates: Pascal Dujardin noted an action to update the architecture draft to clarify how RAW plays with DetNet. This includes scenarios where RAW acts as a superset over DetNet (service layer) or operates at the endpoint (user side of the UNI) for access-specific optimizations.
The Need for a Deterministic Transport Protocol (DTP):
- Pascal argued that current transport protocols (e.g., TCP) are ill-suited for deterministic networks like DetNet and RAW. TCP's windowing and congestion control mechanisms, designed for opportunistic and unreliable transmission, would lead to poor utilization and performance (e.g., dramatic throughput reduction due to discards or ingress queuing) when interacting with DetNet's committed information rates and precise timing.
- He described the need for a transport that is "time-aware," capable of packing data (e.g., 2KB blocks) and delivering them to the network at precise, pre-agreed times (similar to a TDMA streaming protocol). This could involve an interface where the network signals "clear to send" or "pulls" data from the transport layer.
- Pascal drew parallels to historical experiences with SNA networks and early deterministic switches where similar issues arose with window-based protocols.
Relevance to RAW and DetNet Charters:
- Rick (Chair) highlighted that while RAW and DetNet can establish deterministic paths, their utility would be severely limited if transport protocols don't take advantage of these capabilities, potentially leading to near-zero utilization.
- Lou Berger (DetNet Chair) agreed this is a significant, general DetNet problem, not specific to wireless. He suggested that existing transport protocols (like TCP or RTP) could be made "aware" of the underlying network's deterministic capabilities through APIs or modified congestion control algorithms, rather than necessarily requiring an entirely new protocol.
- There was a brief debate between adapting existing protocols (Lou) versus creating a new one (Pascal), with no immediate consensus on the best approach.
Documenting Transport Layer Assumptions/Dependencies:
- Carsten Bormann suggested that if the RAW architecture relies on specific attributes or behaviors from upper/lower layers, these dependencies should be explicitly documented within the architecture.
- Lou Berger further suggested that the broader issue of transport for DetNet should be captured in a standalone DetNet document, rather than being "buried" within the RAW architecture, to ensure wider visibility and discussion within the DetNet WG.
Clarification of "RAW" in Architecture: Lou sought clarification on whether the term "RAW" in the architecture document referred to something entirely new or an evolution/extension of existing IETF technologies (like DetNet and Traffic Engineering). Pascal clarified that RAW is an evolution, building on and extending DetNet's capabilities, particularly for wireless links. The architecture document explicitly reflects this. Rick (Chair) reinforced that the current charter focuses on specific solutions (e.g., PERIO, PSE) and the OODA loop concept, and future broader problems might lead to new charters.

Decisions and Action Items

Pascal Dujardin:
- Update the RAW architecture document to include a few sentences explicitly stating the expectations and dependencies on upper-layer transport protocols for effective utilization of RAW/DetNet capabilities.
- Forward a link to his expired draft-dujardin-tsvtwg-detnet-transport document to the DetNet mailing list. The accompanying email should introduce the document as a potential starting point for a DetNet discussion on transport considerations, highlighting that its first half focuses on problem statements/requirements while the latter half is more solution-oriented.
Rick (Chair): Coordinate an informal follow-up session (potentially remote-friendly for Pascal) to continue discussions on the deterministic transport challenge and how RAW/DetNet should address its interface with the transport layer.
Lou Berger: Provide specific text suggestions to the RAW mailing list for word-smithing the RAW architecture document, particularly concerning the definition and scope of "RAW" (i.e., clarifying its evolutionary nature and relationship to DetNet).

Next Steps

Pascal to execute his action items regarding the RAW architecture document and the DetNet mailing list post.
Rick to initiate coordination for an informal discussion session.
Lou to provide specific text suggestions to the RAW mailing list.
Continued discussion on both the RAW and DetNet mailing lists regarding the deterministic transport problem and the RAW architecture.