Every distance-vector protocol has the same failure mode baked in: a router that loses a route can be re-taught that route — second-hand — by a neighbour who originally learned it from the failed router itself. The result is a routing loop, with the metric slowly 'counting to infinity' as the phantom route bounces between routers, incrementing by one hop per exchange. RIP's loop-prevention toolkit is a set of patches over this flaw, and each one is worth knowing by name because EIGRP inherits the same ideas.
The toolkit
- Split horizon — never advertise a route back out the interface on which it was learned. This kills the trivial two-router loop before it starts and is on by default on almost all interfaces.
- Route poisoning — when a route dies, do not silently drop it; advertise it with metric 16 (infinity) so every neighbour hears about the failure explicitly.
- Poison reverse — a router hearing a poisoned route breaks split horizon deliberately to echo the poison back, confirming 'I know it is dead too'.
- Triggered updates — send an immediate partial update on a topology change rather than waiting up to 30 seconds for the next cycle.
- Holddown timers — after hearing bad news about a route, ignore worse-or-equal news for 180 seconds, so a flapping route cannot whiplash the table.
- Maximum hop count — 15 hops usable, 16 = unreachable, which caps how far counting-to-infinity can run.
Why 16 means infinity
The hop limit is not a size constraint so much as a loop-damage constraint. If a loop forms despite the other mechanisms, the bouncing route's metric climbs by one on each advertisement; at 16 both routers finally agree the destination is unreachable and the loop starves. A higher infinity would mean longer loops; a hop-count infinity of 16 was the compromise. The side effect is brutal: no RIP network can be more than 15 routers deep, period.
The deeper problems
Hop count is a blind metric. RIP will happily prefer a path of two 64 kbps serial links over three 10-gigabit links, because two is less than three. It has no concept of bandwidth, delay or load — the exact gap EIGRP's composite metric was designed to fill.
Convergence is timer-bound. Even with triggered updates, holddown means a failed route typically takes 180+ seconds to be fully replaced — during which traffic blackholes. Link-state protocols like OSPF, where every router holds the full topology and recomputes paths in milliseconds, exist precisely because of this.
Full-table periodic updates waste bandwidth and CPU on links where nothing has changed, and scale linearly with table size.
What to take forward
RIP's obituary is the syllabus for the next two books on this shelf. EIGRP keeps the distance-vector model but replaces the metric, replaces periodic updates with change-driven ones, and replaces counting-to-infinity with the DUAL algorithm's provably loop-free path selection. OSPF abandons distance-vector entirely for link-state flooding. When you study either, you are watching engineers answer, point by point, every weakness this chapter listed.