Update review · November 21, 2026
DAWN Use Cases -02, Four Scenarios Up Front and Two New Authors
What changed in -02?
The two appendices from -01 moved into the body as Section 5, ahead of the taxonomy. Two new scenarios joined them, and each came with a new co-author.
The document went from 26 pages to 31. The -01 review covered the narrower scope and the first two worked examples. Those two examples are unchanged apart from typo fixes.
The header now reads "Discovery of Agents With Names" where -01 said "Individual Submission". The draft name is still draft-kay-dawn-use-cases, so this is not an adopted document.
What is new
- Section 5.3, visitor in an unfamiliar building. Alice joins a building's guest Wi-Fi. Her assistant asks the network which agents are advertised and gets back "Reception" and "Wayfinding". The request names no organisation, entity or capability. DNS-SD over mDNS is given as one way to do it.
- Section 5.4, infrastructure-assisted discovery. A field engineer's assistant connects to a named troubleshooting agent. The enterprise network supplies the requester's site, VPN and tenant. The hosting platform supplies the agent's current endpoint and protocol. These are the two "hooks" from draft-dunbar-dawn-infra-assisted-discovery.
- Scenarios drive the rest of the document. Sections 6, 7 and 8 now cite the scenarios for each characteristic, pattern step and category.
- A new characteristic (Section 6). The starting point of discovery varies: an organisation's name, a known entry point, or only the joined network.
Does the local network scenario hold up?
It is the first scenario with no prior trust relationship. Hermes knows neither the operator's name nor any catalogue it publishes (Section 5.3.1).
Section 5.3.5 handles rogue advertisements well. Guest isolation stops another visitor from advertising an agent called "Reception". The draft says this complements integrity protection and an origin check and does not replace them.
The problem is the origin check. Hermes has no name to check the origin against. The draft never says what an origin check means when the only thing the client knows is the SSID it joined.
There is also a mismatch in the discovery result. Section 5.3.4 says local discovery "adds no fields" to the minimum discovery information. The result it lists is name, protocol and contact details. Entity type and capability identifier, both listed in Section 5.1.5, are missing. Requirements -02 says the minimum MUST include the entity's type and the services it offers (REQ-PROP-1).
The scenario gets around this by having Alice read the names and pick. That works for two agents in a lobby. It does not work for an assistant acting on its own.
Is the infrastructure-assisted scenario discovery?
Barely. An "application or enterprise function" picks the agent before DAWN is involved (Section 5.4.2). DAWN then turns Troubleshooting-Agent.Site-A into an endpoint and a protocol. Requester context only picks the "discovery domain, resolver view, local discovery scope, or bootstrap entry point".
That is name resolution with split-horizon DNS. The scenario does not say what DAWN adds over what enterprises already run.
One sentence needs fixing. Section 5.4.1 says an abstract form of the requester's context "may form part of the minimum discovery information". Everywhere else in the document, that term means what the discovered entity publishes. Requester context is an input.
The scenario also depends on the "Resource State Hook" and "Attachment Context Hook" from an individual draft that has not been adopted.
Was the previous feedback addressed?
Addressed:
- The unfinished media scenario. The "REQ-???" line is gone and the typos are fixed (Section 5.2).
Partly addressed:
- The stale requirements mapping. Both mapping tables were deleted. Section 1 now says the mapping "is maintained in the requirements document, not here". Requirements -02 has no such mapping. It cites this draft once, in its introduction.
- The Discovery Pattern section. Still there as Section 7. A new paragraph walks all four scenarios through the five steps, so it now has a job.
- Folding in single-scenario drafts. The last review asked whether drafts like draft-jadoon-dawn-layered-agent-discovery fold into this one. Two did, by adding their authors. Jadoon's did not.
Not addressed:
- The TBD sections. All five entity class sections are still "TBD" (Section 9). That is three revisions.
- The "[object-based media delivery]" placeholder. Still in the first sentence of Section 5.2.
- Operational Considerations. Unchanged (Section 12).
- Overlap with existing federation and trust work. Still no prior art cited. The only new reference is the Dunbar draft.
Mechanisms added or removed
This is a use case document, so there are no protocol mechanisms. The requirements mapping tables (A.7 and B.5 in -01) are the only removal.
Background of the authors
Two authors were added, for six in total.
Loránd Jakab (Cisco) has 1 RFC, RFC 7215 on LISP deployment. He co-authors the mDNS agent discovery and common discovery parameters drafts. In -01 he was thanked in the acknowledgments for comments on local network discovery. Section 5.3 is that use case.
Linda Dunbar (Futurewei) has 10 RFCs and 11 active drafts, mostly routing and SD-WAN. She is an experienced IETF author. Section 5.4 is a summary of her infrastructure-assisted discovery draft.
Feedback for the authors
- Say what the origin check is in Section 5.3. The client has a network and nothing else. Either name what it verifies against or list this as a known limitation.
- Reconcile Section 5.3.4 with REQ-PROP-1. Either the local result carries type and services, or the minimum differs by deployment setting.
- Justify Section 5.4 or cut it down. State what DAWN provides that a split-horizon resolver does not. Fix the "requester context as minimum discovery information" sentence.
- Put the requirements mapping somewhere. Section 1 points at a document that does not contain it.
- Fill in Section 9 or delete it.