Markdown Version | Session Recording

Session Date/Time: 12 Dec 2022 13:00

DETNET

Summary

This DETNET interim meeting focused primarily on requirements for enhanced and scaling deterministic networking, particularly related to queuing mechanisms and the necessary data plane enhancements. A significant outcome was the confirmation of the working group's intent to consolidate all requirements into a single adopted document. The session also saw presentations of various queuing mechanism proposals, leading to extensive discussion on their underlying principles, metadata requirements, scalability, and encapsulation strategies. A key decision was made to form a design team early next year to guide the technical work in this area, with David Black agreeing to lead.

Key Discussion Points

• Working Group Status:

  • A recent RFC was published, and other documents are nearing completion.
  • One new document adoption, focused on "enhanced detnet data plane requirements," is driving today's discussion.
  • The term "large scale" in draft titles and filenames is being updated to "scaling" to be more inclusive and avoid definitional arguments. This change is expected to propagate to document titles.
  • A liaison response, with recent changes based on mailing list feedback, is pending final comments from the WG before being sent on the 16th.

• Scaling DetNet Requirements (draft-ietf-detnet-scaling-requirements):

  • The adopted draft describes technical requirements for scaling DetNet, addressing heterogeneous characteristics, long hops, numerous flows, and multi-domain scenarios.
  • New requirements include supporting asynchronous and synchronous timers, handling large single-hop propagation latency, higher link speeds, massive traffic, topology changes, and enhanced queuing mechanisms.
  • A new requirement was proposed to accommodate flow fluctuation in aggregated traffic, distinguishing between toleration and recovery/restore mechanisms.
  • Discussion on quantifying "large latency" and "unbounded jitter" in requirements, suggesting examples from industrial and wide-area applications.

• Consolidating Requirements for Enhanced DetNet:

  • Discussion clarified that "enhanced DetNet" is not a separate concept but rather the requirements to improve DetNet for scaling scenarios.
  • There was broad agreement that all "enhanced DetNet" requirements should be incorporated into the existing adopted "scaling" requirements document. Authors of related drafts were encouraged to work with the current document editors to propose text on the mailing list.

• DetNet Queuing Encapsulation and Information Models:

  • Concern was raised about the impracticality of having separate packet headers for each proposed queuing mechanism.
  • A proposal for an information model was presented to define the semantic elements (e.g., sequence numbers, flow IDs, timestamps, experienced delays, deadlines, cycle IDs, priorities) required in packets by different queuing mechanisms. This model would inform packetization in MPLS, IPv6, or BIER without pre-determining specific encoding.
  • The discussion differentiated between fixed ("common") and flexible ("mechanism-specific") parts of metadata.
  • The concept of using DSCP-like code points for configured semantics (rather than hard-coded elements) was introduced, aiming for a total DetNet overhead of around 128 bits in the packet header.
  • A specific proposal for "Boundary Latency Information" (BLI) was presented, classifying it into 'requirement' and 'resource' types, with proposed IPv6 and MPLS extension headers. Debate arose about carrying end-to-end delay budgets in the data plane versus control plane.

• Proposed Queuing Mechanisms:

  • TCQF (Time-Constrained Queuing and Forwarding): Highlighted for its low jitter without requiring new packet headers by leveraging existing fields (e.g., EXP, IP DSCP). Discussion revolved around the architectural decision of using DSCP for DetNet vs. maintaining orthogonality with DiffServ, revisiting an earlier DetNet decision.
  • CQF Variant: A variant of Cyclic Queuing and Forwarding was presented to address limitations of fundamental CQF in large DetNet, particularly for shorter cycle times. It proposes using more than two buffers and introducing "cycle ID" metadata in packets to resolve time ambiguity caused by processing variations.
  • Asynchronous Deterministic Networking (ADN): Proposed solutions to decouple latency and jitter guarantees and avoid strict time synchronization. Mechanisms included Asynchronous Traffic Shapers (ATS), Flow Aggregate and Interleaved Regulators (FAIR), Port-based Flow Aggregate Regulation (P-FAR), and Global Finish Time (GFT) metadata. Scalability of ATS-derived mechanisms and per-flow steering state were discussed as critical considerations.
  • Enhanced EDF (Earliest Deadline First): An enhanced EDF queuing mechanism was proposed for deterministic queues, offering "in-time" (bounded latency) and "on-time" (strict deadline) services. It uses packet information like planned residence time and residence time evaluation, with proposed IPv6 extension headers for DetNet information. Questions were raised about the assumption of "time synchronicity not required" when using absolute deadlines.

• Gap Analysis for Queuing-Based Enhancement:

  • Identified gaps related to supporting aggregated flows and providing deterministic latency for them in large-scale topologies.
  • Emphasized the need to list all proposed queuing mechanisms, determine which require metadata, and design a common, generic format for this information, separating fixed and flexible parts.

Decisions and Action Items

• Requirements Document Consolidation: The working group (WG chairs and AD) supports consolidating all requirements for enhanced/scaling DetNet into the existing adopted working group document (currently draft-ietf-detnet-scaling-requirements).
  • Action Item: Authors with requirements in other drafts (e.g., Jason) are to work with the editors of draft-ietf-detnet-scaling-requirements to propose their contributions to that document on the mailing list.
• Design Team Formation: A design team will be formed early next year to guide the technical work on queuing mechanisms and data plane enhancements.
  • Decision: David Black has agreed to lead the design team.
  • Action Item: The chairs will draft a charter proposal for the design team, informed by discussions in this meeting, and send it to the mailing list, along with a call for participants.
• Liaison Response:
  • Action Item: Review the updated liaison response on the mailing list and provide any final comments by the 16th.

Next Steps

• Mailing List Discussions: Continue technical discussions on the DetNet mailing list for all proposed queuing mechanisms, metadata requirements, and encapsulation strategies.
• Design Team Kick-off: Once formed, the design team will begin work on the consolidated requirements document and developing solutions for DetNet queuing. Design team meetings will be open to all.
• Author Collaboration: Authors of similar or overlapping drafts (e.g., queuing mechanisms, metadata proposals) are encouraged to meet informally and collaborate ahead of the design team's formal start. The chairs can facilitate setting up informal meeting resources.
• Clarifications on Specific Drafts:
  • draft-ietf-detnet-scaling-requirements: The updated version addressing comments (including flow fluctuation) should be posted to the mailing list.
  • DSCP Use: The question of whether DetNet should use DSCP fields or maintain architectural orthogonality with DiffServ will be a topic for the design team.
  • Time Synchronization: Clarification is needed on the assumptions regarding time synchronization for deadline-based queuing mechanisms.
  • Metadata Information Model: Further work on defining semantic information elements for packet headers, including their properties (rewriteability, bit requirements).
  • Queuing Mechanism Prioritization: The working group will need to confirm which queuing mechanisms should be prioritized for enhancement within DetNet.

Related Documents

draft-ietf-detnet-scaling-requirements