End-to-End Delivery Timeouts for MOQT

draft-sharma-moq-end-to-end-delivery-timeout-00 | Individual submission | -00 posted 2026-10-06 | 6 pages | expires 2027-04-09 | Datatracker

Authors

• Aman Sharma (Meta)

Abstract

A standards-track moq-transport extension that lets objects carry a total delivery deadline measured from the original source, rather than a per-hop budget that resets at every relay. MOQT’s existing OBJECT_DELIVERY_TIMEOUT / SUBGROUP_DELIVERY_TIMEOUT restart at each relay, so a single object can accumulate a fresh timeout at every hop and stale content can pile up delay across a multi-relay path without ever being dropped. This draft adds an end-to-end timeout that tracks the elapsed time since the publisher first forwarded the object, so any relay (or the final subscriber) can discard content that is already too old overall.

Sharma posted it 2026-10-06 in the “Timestamp Draft” thread, noting it is “based on your TIMESTAMP draft and inspired by the discussion Martin started earlier” — i.e. it builds on the timestamp-as-properties work and on Martin Duke’s delivery-timeout framing.

Key Technical Details

• END_TO_END_DELIVERY_TIMEOUT Setup Option — negotiates the extension at session setup.
• END_TO_END_OBJECT_DELIVERY_TIMEOUT Message Parameter — a varint in milliseconds, carried on SUBSCRIBE, SUBSCRIBE_TRACKS, PUBLISH, or REQUEST_UPDATE.
• Reference timestamp / reference time — the publisher records the timestamp and local wall-clock time of the first forwarded object and uses them as the baseline. An object’s deadline is then computed as deadline = reference_time + (T − reference_timestamp) / timescale + timeout, where T is the object’s timestamp. No synchronized clocks are required between endpoints — each hop reasons about accumulated delay from the same reference pair.
• Independent of the per-hop timeouts — the draft states this timeout is independent of OBJECT_DELIVERY_TIMEOUT and SUBGROUP_DELIVERY_TIMEOUT; the first applicable timeout to expire determines the result.

Relationship to other work

• Builds on the timestamp drafts. It consumes the object timestamp / timescale that [[moq-timestamp-properties|draft-frindell-moq-timestamp]] and [[moq-timestamp|draft-lcurley-moq-timestamp]] define as properties — a concrete example of the “factor timestamps out as a reusable property layer, then build semantics on top” approach Frindell described on-list, and of the layering Luke Curley endorsed in the same thread.
• Delivery-timeout lineage. Directly continues Martin Duke’s long-running “timestamps solve delivery timeout” argument (the motivation behind moq-timestamp in June 2026) — the gap being that per-hop timeouts can’t bound total source-to-sink age.
• Implementation echo. quiche moqt split delivery_timeout into separate subgroup and object delivery timeouts (commit 2026-10-06), refining the per-hop timeout machinery this draft layers an end-to-end bound on top of.

Notes

Individual submission by Aman Sharma (Meta); not WG-adopted. Posted 2026-10-06, the newest MoQ document on the datatracker, days before the Oct 12–15 Seattle hybrid interim. See moq-timestamp-properties, moq-timestamp, moq-transport, discussions-2026-10.

Links

• Datatracker: https://datatracker.ietf.org/doc/draft-sharma-moq-end-to-end-delivery-timeout/
• Discussion: moq@ietf.org mailing list, “Timestamp Draft” thread (permalink, announcement 2026-10-06)