Session Date/Time: 07 Jun 2023 00:00

DETNET

Summary

This DETNET interim meeting featured two main topics: an initial evaluation of the Cyclic Queuing and Forwarding (CQF) mechanism (IEEE 802.1Qch) against the DETNET scaling requirements draft, and a detailed presentation of a new flexible time slot mechanism proposal. The CQF evaluation led to significant discussion and identified areas for clarification and refinement in the requirements document itself. The new time slot mechanism presentation prompted questions regarding its unique contributions and distinctions from existing DETNET proposals.

Key Discussion Points

Initial CQF Evaluation against Scaling Requirements (Izu)
- Izu presented a preliminary evaluation of the published IEEE 802.1Qch CQF standard against eight requirements from the DETNET scaling requirements draft.
- Tolerate Time Asynchronously: CQF was rated 'No' as it mandates time synchronization.
- Support Large Single-Hop Propagation Delay: CQF was rated 'No'. Large propagation delays necessitate very large cycle intervals (TC) to maintain utilization, which in turn leads to large end-to-end latencies.
- Accommodate Higher Link Speed: CQF was rated 'Partial'. The original CQF was not designed or tested for speeds higher than 1Gbps. Higher link speeds don't inherently improve clock accuracy or processing latency variation, making it harder to determine packet cycles.
- Scalable to Large Number of Flows: CQF was rated 'Partial'. Transmission control via traffic classes is scalable (e.g., 8 classes, two buffers use two classes). However, stream gate filtering (PSFP) at input ports, if applied per-stream, could limit scalability; aggregation is more common.
- Tolerate High Utilization: CQF was rated 'No'. Not all cycle time is usable for real data traffic due to dead time. For small cycle times (tens of microseconds), achieving high utilization is difficult. The discussion also highlighted ambiguity in defining "high utilization" within the requirements document (e.g., utilization of reserved bandwidth vs. full link bandwidth, impact of propagation delay).
- Prevent Flow Fluctuation from Disrupting Service: CQF was rated 'Partial'. For compliant flows, it works. For non-compliant flows, limited buffers (typically two) can lead to packet drops if a packet misses its intended buffer's cycle. Discussion ensued on whether the requirements document should explicitly address non-compliant flows or if they are outside the scope of guaranteed services. A sense of those present indicated that this requirement, when referring to burst accumulation, is met by CQF.
- Scalable to Large Number of Hops with Complex Topology: CQF was rated 'No'. For a fixed bounded end-to-end latency, a large number of hops (e.g., 100) becomes problematic because cycle time (TC) must be extremely small. However, jitter bound (approx. 2*TC) remains acceptable. This point also led to a discussion on how "scalable" is defined: whether end-to-end latency is fixed, or if it can scale with the number of hops. The current text in requirements draft 3.7 is considered very brief, needing expansion.
- Overall: The evaluation generated significant discussion, highlighting areas where the DETNET requirements draft needs clearer definitions (e.g., "high utilization," "scalability," handling of non-compliant flows) to enable unambiguous evaluation of mechanisms.
Detailed Presentation on Flexible Time Slot Mechanism (Xiaofu)
- Xiaofu presented a new time slot mechanism aimed at providing more flexible resource allocation than traditional CQF, improving service scale, and reducing complexity.
- The proposal is inspired by CQF but seeks to overcome limitations such as reduced service scale or complexity in multi-buffer/multi-performance modes.
- It introduces a time-slot model for forwarding resources, where resources are available to services based on "M-unit multiples" of the actual device capacity (M).
- The core steps involve determining outgoing sending time slots, reserving these slots via a control plane, constructing time-slot-typed forwarding resources on links, and accessing these slots in the data plane.
- A "global time slot mapping" concept was proposed to simplify state maintenance across hops.
- Discussion: There were questions about how this proposal differentiates from and improves upon existing DETNET proposals, particularly scQf (Streaming-capable CQF), which has been presented previously in the WG. Clarification was requested on the specific problems this mechanism solves, its detailed latency calculation, and how it delivers on "flexibility." Concerns were also raised about the complexity of admission control and the scalability of maintaining states for "sub-bursts."

Decisions and Action Items

ACTION: Izu to refine the CQF evaluation, incorporating feedback from the discussion, and collaborate with the DETNET requirements draft authors (Ping and co-authors) to suggest concrete changes and clarifications for the requirements document, particularly concerning "high utilization," "flow fluctuation," and "scalability to large number of hops." (Deadline: To be discussed on mailing list).
ACTION: Xiaofu to revise the flexible time slot mechanism draft. The revision should clearly articulate the problem(s) it solves that are not addressed by existing DETNET mechanisms (e.g., scQf), provide a detailed comparison to relevant proposals, and elaborate on the specific benefits, such as how it improves flexibility and calculates latency/jitter. (Deadline: To be discussed on mailing list).
ACTION: (General encouragement) Other participants are encouraged to conduct similar evaluations of DETNET mechanisms against the requirements draft to help identify further areas for clarification and refinement in the requirements document.

Next Steps

Continued discussion on the DETNET mailing list regarding the requirements draft refinements and the flexible time slot mechanism proposal.
Await revised versions of the requirements and flexible time slot mechanism drafts based on the discussions.