Unfortunately, the world of IP routing can be overwhelming, requiring detailed knowledge of multiple intricate protocols. Although many good books, courses and training programs are available for in-depth study (there is a cottage industry in support of IP routing), we will describe the fundamental principles for two of the most common routing protocols found on corporate networks: RIP (Routing Information Protocol) and OSPF (Open Shortest Path First).

RIP was originally distributed along with Berkeley's Unix, and like many of the other BSD services, it has become a critical element of IP networks everywhere, even though it was not developed as an Internet standard. Two distinct versions of RIP are now documented as IETF protocols: RFC 1058 describes the original RIP (version 1), while RIP v2 is defined by RFC 1722 (also published as Internet Standard 56). Although the protocols share many similarities, there are some important differences between them.

RIP uses a "distance-vector" algorithm that associates a specific "distance" (the number of hops) with a specific "vector" (a destination network or host). RIP devices learn about destinations and their distance from neighboring RIP routers, then choose the path to a destination based on the route with the lowest number of hops. Once a route for a destination has been chosen, it is stored in a local database, and any other routes for that destination are discarded. Periodically, each router advertises all the paths it has discovered. Eventually, all the devices on a network will discover the best routes to all the available destinations.

RIP defines hops as the number of routers between a sender and the destination network or system. If a router is attached to a network, the distance to that network is zero hops. Similarly, if a router can reach a neighboring network only by sending the datagrams to a neighboring router, the distance to that vector will be one hop. Whenever a router advertises a route, it increments the known distance metric by one hop. As these broadcasts arrive at the neighboring routers, they are compared with the entries already in those routers' databases. If one of the advertised routes to a destination is shorter than an existing entry, the advertised route will be incorporated into the local routing table, with the advertising router being listed as the next hop for the destination.

Some of these issues are clarified in "Enterprise Routing Model" (below), which shows five different network segments. In that example, Router A would advertise single-hop routes for the Ethernet segments and Internet links to which it was directly attached. Meanwhile, Router B would advertise single-hop routes for its local Ethernet attachment, a single-hop route for the WAN subnet and multihop routes for the networks in the remote areas (including the network segment that Server Z is attached to). All these routes will be advertised with broadcasts every 30 seconds, and both routers will republish the routes that they learn from each other.

In the RIP distance-vector model, this means that Router B would see the distance to Server Z's segment as being one hop through Router A, while Router A would see the WAN network as one hop through Router B. However, both routers would see their shared Ethernet as zero hops from themselves, and one hop from the other router. Because the advertised route for that segment is longer than the known route, the advertisements for the shared segment would be discarded by both routers.

Enterprise Routing Model Click here to enlarge

RIP Damage Control

RIP imposes a maximum hop count of 15, and any route that is advertised with a distance of 16 hops must be considered as unreachable through the advertised path. RIP uses an expiration timer of 180 seconds. If a route goes silent for three full minutes, the other routers that have learned about the route will set the hop count for the path to 16 (thereby flagging the route as invalid) and will advertise the route with the new distance value for another 120 seconds to ensure that downstream systems learn about the failure.

However, an older route (with the legacy distance metric) may still be advertised by a distant router. Since that route would have a lower value than that of the infinite path, the remote route would propagate back through the network. For example, if a failure caused Router A to remove Server Z's network segment from its local routing table, then the next update from Router B (with a hop count of 2) could be incorporated into Router A's database as the shortest distance to that destination. In effect, Router A would see Router B as the shortest distance to the Server Z segment and would send traffic for that segment to Router B, though Router B would send it right back to Router A.

One technology that RIP uses to prevent these problems is called split horizon, which prohibits routers from advertising routes through the same interface from which the route was learned. In our example, split horizon would prevent Router B from advertising Router A's local connections back to the Ethernet segment, and vice versa. With this technique, Router A may lose sight of the Server Z segment, but it would never learn of a better path through Router B.

Another technique often used to prevent localized routing loops is poison reverse, which uses a hop count of 16 to explicitly corrupt the reverse path. In that case, Router B would advertise Server Z's network segment with a distance of 16 hops when sending updates, ensuring that the route was not used by Router A. Note that neither of these mechanisms prevents routing loops from occurring when multihop loops are designed into the network but instead prevents only localized loops from occurring through normal operations.