Session Date/Time: 26 Apr 2023 12:00

DETNET

Summary

The DETNET Working Group meeting focused on two main areas: further refinement of the "Enhanced Data Plane Requirements" draft (draft-ietf-detnet-enhanced-data-plane-requirements) and a presentation on "Deadline Based Forwarding" (draft-li-detnet-deadline-based-forwarding). Discussion on the requirements draft centered on clarifying new proposed requirements, separating data plane from control/management plane concerns, and ensuring compatibility with various forwarding mechanisms. The deadline-based forwarding presentation introduced a queuing and scheduling mechanism based on an Earliest Deadline First (EDF) variant, followed by a detailed Q&A session covering its operation, assumptions, and potential issues like packet reordering.

Key Discussion Points

Meeting Process: The chairs noted that having one presentation per meeting allows for more thorough discussion, a practice that will continue.
Requirements Draft (draft-ietf-detnet-enhanced-data-plane-requirements):
- New Proposed Requirements: Jindong proposed three new requirements for Section 3: handling a large number of hops, tolerating burst accumulation, and tolerating high utilization.
- Requirement 3.7 Clarification: Discussion arose on whether two paragraphs within the current requirement 3.7 should be considered one or two distinct requirements.
- Requirement 3.8 Split: A sense of those present indicated that the proposed requirement 3.8 ("tolerate high utilization and scalable network planning and queuing solutions") contained two separate concerns: data plane aspects related to queuing solutions supporting high utilization, and control/management plane aspects related to network planning. It was suggested that network planning material might be better suited for the detnet control plane framework draft (draft-ietf-detnet-control-plane-framework).
- Requirement 4 (Path Selection): Discussion clarified that this requirement should focus on the data plane's compatibility with various steering solutions (e.g., source routing, hop-by-hop steering) and the interaction of multiple extension headers (e.g., for steering and QoS). The objective is for queuing mechanisms to be applicable to whatever steering is in use, rather than being specific to one.
- Header Design Implications: Participants recognized the need to consider what per-packet information queuing mechanisms require to be carried in headers, without prematurely committing to specific header designs.
Evaluation of Queuing Mechanisms against Requirements:
- Initial Evaluation Target: The chairs proposed that for the next meeting, the requirements draft authors begin an initial evaluation of a known queuing/scheduling mechanism against the detnet requirements.
- TSN ATS vs. RFC 2211: There was a brief discussion on whether to use RFC 2211 (InterServ) or TSN ATS (Time-Aware Shaper/Scheduler) for this initial evaluation. While RFC 2211 is an IETF RFC, TSN ATS was generally preferred as a more relevant and modern mechanism, despite potential challenges in accessing IEEE standards documents (though it was noted that many IEEE standards are free after six months).
Deadline-Based Forwarding Presentation (Xiaofu Li):
- Mechanism Overview: The presentation introduced a queuing mechanism that uses an Earliest Deadline First (EDF) variant, maintaining queues based on a "countdown time" (CT) and a "rotation interval" (TI). It described "in-time" (work-conserving) and "on-time" (non-work-conserving, delaying packets until closer to their deadline) modes.
- Schedulability and Bursts: The draft provides schedulability conditions and discusses how bursty traffic is handled through latency compensation, which acts as a shaper at ingress/transit nodes.
- Buffer Sizing and Admission Control: The approach considers buffer sizing based on schedulability conditions and relies on admission control at ingress to ensure traffic adheres to its "contribute function."
- Packet Header Fields: The mechanism relies on packet headers carrying "plant earliest sending time" (D) and "accumulated residence time deviation" (E) values, which are updated per hop. D is calculated by a controller, while E is updated by each node.
- Clock Assumptions: The mechanism does not require global wall clock time synchronization but assumes that router clocks are frequency-synchronized, meaning their rates of advance are aligned (bounded clock skew).
- Packet Reordering: It was acknowledged that latency compensation and work-conserving behavior could lead to packet reordering within a flow. Two potential solutions were mentioned: "packet ordering focusing" (from DetNet) or using a "flow/class object" for reordering.
- Comparison to RC EDF: A participant asked for clarification on the difference between this proposal and Rate-Controlled Earliest Deadline First (RC EDF), noting similarities in using shaping (latency compensation) and EDF scheduling. The presenter clarified differences in per-flow state and queue management.
- Calculus Proof: The proof for the calculus used in the mechanism is based on existing literature referenced in the draft.
- Header Field Size: A suggestion for the size of the deadline field in a header was 3 bytes, consistent with some existing IPv6 extension header practices, to accommodate potential long paths and large deadlines.
- "On-time" Mode: It was clarified that the "on-time" (non-work-conserving) mode, while offering higher priority, does not necessarily guarantee latency bounds and is primarily intended for ingress nodes, not transit.
- IOAM Analogy: It was noted that IOAM (In-situ OAM) mechanisms also update per-packet fields at each hop, which might offer insights into practical implementations of updating fields like 'E'.

Decisions and Action Items

Decisions:

The working group will adopt a practice of focusing on one presentation per meeting to allow for more in-depth discussion and Q&A.
Initial evaluation of queuing mechanisms against the detnet requirements will prioritize TSN ATS over RFC 2211 due to its modern relevance.

Action Items:

Requirements Draft Authors (Peng, Jindong, et al.):
- Continue discussions on the proposed changes to requirements (especially 3.5, 3.7, 3.8, and 4) via the mailing list.
- Revise the draft-ietf-detnet-enhanced-data-plane-requirements based on these discussions, clarifying the separation of data plane vs. control/management plane requirements.
- For the next meeting, prepare an initial draft matrix evaluating TSN ATS against the requirements specified in the requirements draft.
Xiaofu Li (Deadline-Based Forwarding): Continue to refine the draft-li-detnet-deadline-based-forwarding, particularly addressing the packet reordering issue and providing more detailed descriptions of the calculus and header field usage.
Chairs: Upload the presentation slides to the working group's document repository, ideally with slide numbers added.
All Participants: Review the meeting notes in HedgeDoc for accuracy and provide feedback.

Next Steps

Mailing list discussions will continue on the proposed revisions and clarifications for the "Enhanced Data Plane Requirements" draft.
The requirements draft authors will work on preparing an initial evaluation of TSN ATS for the next meeting.
Work on the deadline-based forwarding draft will continue, with an emphasis on addressing the identified technical considerations.
The next working group meeting is planned for approximately two weeks from now.