Count-to-Infinity in Distance-Vector Routing & Countermeasures (Split Horizon, Poison Reverse)

Distance-vector routing (DVR) protocols compute routes using iterative neighbor exchanges. A key limitation is that they do not maintain full topology knowledge; instead, routers infer reachability from the distance vectors received from neighbors. As a result, when a destination becomes unreachable, routers can temporarily “believe” the destination is reachable via each other, creating a routing loop that causes metrics to increase step-by-step—this is the count-to-infinity problem.2

The core mechanism behind the count-to-infinity problem is uncertainty: if router $B$ hears from neighbor $A$ that $A$ can reach destination $D$, then $B$ cannot distinguish whether that path from $A$ to $D$ relies on $B$ itself. When the direct path to $D$ fails, routers may start reinforcing each other’s belief in a broken route, so the hop-count (or cost) estimates keep incrementing until a bounded “infinity” threshold halts the process.2

Key takeaways:

• Distance-vector routing uses iterative updates and relaxations (Bellman-Ford style), which can allow loops.
• Routing loop triggers count-to-infinity.
• Protocols use bounded metrics (an “infinity” value) so the loop cannot grow without limit.

Terminology and the intuition (why metrics increase)

In DVR, each router maintains a table with an estimate for the “distance” to each destination, typically based on the neighbor’s advertised distance plus a local link metric. When a failure happens, routers must retract or update their estimates. But because they only know neighbors’ advertised distances (not the actual path), a router can accept an alternative that is actually looping back through itself.

The typical phenomenon:

1. A path to destination $D$ breaks.
2. One router (say $B$) eventually stops receiving “good news” and increases/removes its entry.
3. Before convergence completes, neighbors still advertise stale information; these update cycles can create a loop.
4. The measured distance estimate increases each round (“one more hop each time”) until reaching a defined infinity threshold.2

In RIP specifically, hop-count metrics are limited: valid distances are $1$ through $15$, and a metric of 16 indicates infinity/unreachable. Therefore, when routing loops cause repeated increases, the process bottoms out at the infinity bound.

Key terms for this mechanism:

• Bellman–Ford relaxation
• Stale update
• Infinity metric
• Convergence

Why Split Horizon helps (but doesn’t fully solve count-to-infinity)

Split horizon reduces the chance that a neighbor will incorrectly infer a loop using reverse knowledge from the same interface. The basic rule is: if a router learns a route to destination $D$ via neighbor/interface $X$, it will not advertise that same route back to $X`—so the most direct feedback loop (often a 2-router loop) is suppressed.

In practice:

• When a route fails, routers can still exchange bad/stale information through other paths.
• Split horizon specifically prevents advertising back on the same interface, so loops longer than 2 hops can still arise, meaning split horizon mitigates but does not eliminate count-to-infinity in all cases.2

Split Horizon: concrete effect on DV loops

For a router $R$ and destination $D$:

• If neighbor $N$ is the next-hop learned route for $D$ on interface/link $i$,
• then $R$ omits advertising the $D$ route to $N$ on $i$.

This reduces “echoes” in routing announcements that commonly start loop formation. However, as noted in discussions of the count-to-infinity phenomenon, multi-router loops (length $\ge 3$) may survive split horizon filtering, so the metric can still climb until infinity.

Poison Reverse: more aggressive than Split Horizon

Poison reverse improves loop avoidance by not merely omitting advertisements but explicitly poisoning them (advertising infinity) back to the neighbor that provided the route.2

Compared to split horizon:

• Split horizon: “I won’t tell you this route on that interface.”
• Poison reverse: “Even if you think I have a route, I’m telling you it’s unreachable via me (advertise infinity).”

Operationally in RIP terms:

• When a router learns a route to destination $D$ from neighbor $N$,
• it sends advertisements back to $N$ with metric 16 (infinity), indicating that $N$ should not route to $D through this router.2

Poison reverse is designed to prevent the neighbor from using the poisoned route as a basis for a looping alternative, thereby helping convergence behavior and loop elimination in more scenarios than simple split horizon.2