Summary

  • RFC 3316 asked cellular hosts to reuse TCP acknowledgments, RTCP reception reports and SIP responses as positive evidence that recent outbound traffic had crossed the first-hop router.
  • The inference was deliberately narrow: forward progress to a remote peer could mark a neighbor recently reachable and suppress a probe, but it did not prove application completion, lasting connectivity or even that arbitrary inbound traffic validated the forward path.

The costly packet in an early cellular IPv6 session was not always payload. Sometimes it was a question the host already had enough evidence to answer.

IPv6 Neighbor Unreachability Detection, or NUD, tracks whether a node’s next hop has recently been reachable. When that confidence expires and traffic is active, the host can send a unicast Neighbor Solicitation and wait for a solicited Neighbor Advertisement. On Ethernet, the exchange belongs naturally beside address resolution. The GPRS and UMTS link described by RFC 3316 looked different. It resembled a point-to-point link: the cellular host had one default router on that interface, and the link did not use link-layer addresses. Address resolution was unnecessary, yet reachability still mattered.

A failed first hop could leave packets moving into a dead path.

The profile’s answer was to ask a sharper question: had some other protocol already demonstrated that a recent outbound packet went forward?

The generic Neighbor Discovery rule supplied the logic. A newly received TCP acknowledgment can show that previously sent data reached the remote peer. For an off-link destination, that packet could not have reached the peer without first passing through the sender’s next-hop router. The end-to-end acknowledgment therefore contains a local fact: the first hop was reachable when that recent packet crossed it. The host may feed that fact into its Neighbor Cache instead of spending another exchange to probe the router.

RFC 3316 carried the same reasoning into the UDP-heavy cellular applications of its time. UDP itself provides no delivery acknowledgment, so the application had to expose a signal with the right causal meaning. For RTP media, an RTCP reception report block indicating that some packets had arrived at the peer could serve. For SIP, a response to a request showed that the request travelled outward. A cellular host acting as a SIP server generally lacked that evidence merely by sending a response, but a later SIP ACK could show that the response had reached the other side.

These examples were not interchangeable status lights. Each had to imply delivery of recently sent traffic. An unrelated incoming datagram did not qualify. Nor did an unsolicited Router Advertisement. That advertisement showed only that a packet could travel from the router to the host. NUD needed confidence in the opposite direction—from the host through its next hop. The distinction is the difference between observing activity and proving the path under test.

The resulting state was also temporary. Upper-layer confirmation could move a Neighbor Cache entry to REACHABLE for a bounded interval. If fresh confirmation stopped, the entry could age to STALE; sending traffic could move it through DELAY and eventually PROBE, where dedicated Neighbor Solicitations resumed. RFC 3316 did not abolish NUD. It let the host avoid asking a redundant question while current application traffic was already supplying a valid answer.

That mattered because radio signaling was not free. The document described cellular bandwidth as constrained and airtime as potentially billable. Eliminating needless messages could protect both capacity and the user’s bill. Yet RFC 3316 reported no measured packet saving, battery gain or charge reduction. It offered an implementation profile built from the link’s topology and the evidence already present in active flows.

The same mechanism survived the document. RFC 7066 replaced RFC 3316 in 2013, updated the network model to include EPS, retained TCP, RTCP and SIP examples, and added DNS responses as another possible application-level confirmation. The later text’s continuity shows that the inference remained useful in the profile. It does not prove which devices implemented it or how often it suppressed a probe.

The historical lesson is not that applications should decide routing. It is that a state machine may safely reuse evidence from another layer only when the implication is explicit. A remote peer’s report can say something about the first hop because the recent outbound packet had to cross that hop. But the conclusion must stop there. An RTCP report does not prove good media; a SIP response does not prove a completed call; a reachable router does not prove a reachable service; and evidence about one packet does not certify the next one.

Sources: RFC 3316, RFC 2461, RFC 4861, RFC 7066, RFC 7849, RFC 8504, RFC 3550, and RFC 3261.