Junos Router | Stability and performance tweaks

Hold downs serve to increase stability, at the expense of rapid convergence, by preventing installation of a route with a reachable metric, after that same route was recently marked as unreachable (cost = 16) by the local router. This behavior helps to prevent loops by keeping the local router from installing route information for a route that was originally advertised by the local router, and which is now being readvertised by another neighbor. It's assumed that the slow count to infinity will complete before the hold down expires, after which the router will be able to install the route using the lowest advertised cost.

Split horizon prevents the advertisement of routing information back over the interface from which it was learned, and poisoned reverse alters this rule to allow readvertisement back out the learning interface, as long as the cost is explicitly set to infinity: a case of "I can reach this destination, NOT!" This helps to avoid loops by making it clear to any receiving routers that they should not use the advertising router as a next hop for the prefix in question. This behavior is designed to avoid the need for a slow count to infinity that might otherwise occur because the explicit indication that "I cannot reach destination X" is less likely to lead to misunderstandings when compared to the absence of information associated with split horizon. To prevent unnecessary bandwidth waste that stems from bothering to advertise a prefix that you cannot reach, most RIP implementations use split horizon, except when a route is marked as unreachable, at which point it is advertised with a poisoned metric for some number of update intervals (typically three).

Triggered updates allow a router to generate event-driven as well as ongoing periodic updates, serving to expedite the rate of convergence as changes propagate quickly. When combined with hold downs and split horizon, a RIP network can be said to receive bad news fast while good news travels slow.