Summary

  • RFC 1256’s default advertisement lifetime was 1,800 seconds: a host’s maximum waiting period before forgetting an unrefreshed, dynamically learned router—not a promise that traffic would fail over within 30 minutes.
  • The RFC explicitly says its default 7.5–10-minute advertisement cadence is too slow to detect a first-hop black hole before a transport session times out; RFC 1122 assigns dead-gateway detection to a separate host function.

The unsettling part of a failed gateway is not always the failure itself. It is the interval in which the host still believes the gateway belongs on its list. That belief can outlive the router’s ability to forward.

In September 1991, RFC 1256 offered hosts a way to learn neighboring routers without a manually maintained address list or dependence on a particular routing protocol. Routers sent ICMP Router Advertisements; hosts could also send a small number of Router Solicitations when an interface started. The scheme made membership discoverable. Its most revealing design choice was how long that membership should remain believable.

The defaults were deliberately quiet. A router’s maximum advertisement interval was 600 seconds, while the minimum default was 75 percent of that: 450 seconds. Each next interval was randomized within the range, so ordinary advertisements arrived roughly every 7.5 to 10 minutes rather than in lockstep. The advertised lifetime defaulted to three times the maximum interval: 1,800 seconds, or half an hour.

That lifetime was a lease on a piece of local state. When a host accepted an advertisement for a neighboring router, it added the address to its default-router list and started a timer from the advertised value. Further advertisements refreshed the timer. If the timer ran out, the dynamically learned entry was removed. A statically configured router was different: it did not acquire this expiration timer merely by being configured.

It is tempting to read “30 minutes” as a failover setting. The RFC says the opposite. It explains that the default advertisement rate and lifetime keep overhead low even when many routers share a link, but that the defaults are not sufficient to detect a first-hop black hole before a transport session times out. Operators could configure shorter values. The standard does not say they did.

The distinction matters because discovery and liveness answer different questions. An advertisement says, in effect, “this address is offered as a neighboring router, with this preference, for no longer than this interval without a refresh.” It does not test whether a packet to a particular destination can traverse the router now. RFC 1256 says it is not a routing protocol: it does not choose the best route for every destination. Redirects handle a different question—whether a better first hop exists for a destination already sent through a router.

RFC 1122 makes the other half of the architecture explicit. The IP layer must detect a failed next-hop gateway and select an alternate, but in 1989 the standards authors said no fully satisfactory general algorithm had emerged. Continuous pinging was ruled out as expensive and unscalable. Instead, higher and lower layers could advise IP: an acknowledged TCP segment is positive evidence; repeated transport failure or a link-layer signal may be negative evidence. The timer in RFC 1256 was never meant to replace that judgment.

Startup solicitation offered a faster path to initial knowledge, not a permanent heartbeat. A host could send at most three solicitations, three seconds apart, and had to stop after a valid advertisement arrived. Thereafter, periodic advertisements kept the list fresh. Silence eventually erased an entry; it did not, by itself, establish when the path had become unusable.

The historical design is therefore less about a half-hour outage than about separating two kinds of evidence. Router Discovery economized on configuration and kept a bounded memory of who had announced themselves. Dead-gateway detection had to reason about current forwarding, traffic, and signals from other layers. Reading the expiration timer as a service-recovery promise collapses those jobs—and attributes a guarantee the RFC explicitly declined to make.

The evidence is in RFC 1256 and the host requirements in RFC 1122, §3.3.1.4. They establish protocol defaults and design boundaries, not how quickly a named operating system, network, or user recovered from a real gateway failure.