Hi all, From my perspective, Option 2 (Generic Metadata Flag) seems preferable to Option 1 and Option 3. This solution avoids dedicating a scarce SRH flag bit to a single function, while allowing the same mechanism to be reused for PSID and potential future extensions. Using an opcode-based approach (assuming it refers to the SID function) also appears to require relatively limited changes to the SRH processing mechanism compared to flag-less alternatives, achieving a better balance between processing efficiency and operational complexity.
Best regards, Lijie Deng From: zengguanming Date: 2026-01-23 16:16 To: SPRING WG List CC: Cheng Li; [email protected]; DHRUV DHODY; chengweiqiang; [email protected] Subject: [spring] Seeking WG Consensus on PSID Encoding Options for draft-ietf-spring-srv6-path-segment Dear SPRING WG, As part of our ongoing effort to finalize the encoding mechanism for the SRv6 Path Segment Identifier (PSID) in https://datatracker.ietf.org/doc/draft-ietf-spring-srv6-path-segment/, we would like to present three high-level approaches—along with their sub-options—for community review and consensus. Thanks to Bruno’s constructive review, comments and thorough discussion, we finally come up with the following options and present to the WG: Option 1: Dedicated P-flag (Current Draft Approach) Mechanism: Introduce a new SRH flag (e.g., P-flag) solely to indicate that SRH. SegmentList[Last Entry] carries a PSID. Pros: Simple, unambiguous, and enables per-packet fast-path processing for precise OAM (e.g., loss measurement). Cons: Consumes one of only eight SRH flags for a single function. Option 2: Generic Metadata Flag (Recommended Evolution) Mechanism: Define a generic SRH flag (e.g., G-flag) that signals the presence of a structured 128-bit sid in SegmentList[Last Entry]. The opcode is defined to distinguish different use cases, for example: OpCode=0x01: Path Segment ID (PSID) OpCode=0x02: In-situ OAM trace data OpCode=0x03: Custom telemetry payload Pros: One generic flag supports multiple future extensions, thus addresses “resource waste” concern by making the flag generically useful. Maintains high-performance, per-packet processing. Cons: Slightly more complex: requires defining opcode semantics and extensibility model. Option 3: No New Flag This has three sub-options: 3A: Reuse O-flag Mechanism: Use the existing OAM flag to signal PSID presence. Pros: • No SRH flags consumption. Cons: • O-flag implies slow-path, sampled OAM treatment (per RFC 8754), but PSID often requires fast-path, per-packet handling for accurate end-to-end metrics. Mismatch in processing model risks under-serving key use cases. 3B: Flag-less (Pure SID Convention) Mechanism: Rely solely on the END.PSID behavior code (Function = 0x0064); no flag needed. PSID is placed at SegmentList[n] where n = SRH.LastEntry. Pros: Minimalist design; No SRH flags consumption. Cons: No visibility for intermediate nodes—limits future telemetry or policy enforcement. Functionally restricted to egress-only use cases (e.g., basic path binding), losing the full programmability advantage of SRv6. 3C: Flag-less with Dedicated PSID Prefix Mechanism: Reserve a well-known, non-routable IPv6 prefix (e.g., ::/32) for PSIDs. Intermediate SR Endpoint nodes inspect SegmentList[n] and recognize PSID by prefix match. Pros: No SRH flag consumption. Enables intermediate node visibility without a flag. Cons: SR nodes on the path needs one more mechanism to read PSID at Segment List[n], which introduces more complexity Next Steps We believe Option 1(Dedicated P-flag) is simple, unambiguous, and enables per-packet fast-path processing for precise OAM, and Option 2 (Generic Flag) offers the best long-term balance: it conserves scarce flag space, supports future extensions (beyond PSID), and maintains performance. And we kindly ask the WG to share your views on: Which direction best meets operational and architectural needs? Any strong objections to the proposed options. Depending on feedback, we will update the draft accordingly and aim to request WGLC soon. Thank you for your engagement! Best regards, Guanming Zeng & Cheng Li Huawei
_______________________________________________ spring mailing list -- [email protected] To unsubscribe send an email to [email protected]
