Routing

A packet crosses several routers on its way, and every one has to answer the same question: where do I send this next? The answer comes from the router's routing table, a list of destinations with the next hop to use for each and usually a cost (a measure such as delay or number of hops). The router reads the destination IP address, finds the best matching entry, and passes the packet on. If nothing matches, it uses the default route (“send everything else to the provider”) or drops the packet. The question is how the tables get filled in.

With static routing a network administrator types the routes in by hand: the destination network, the mask and the next hop. The table never changes unless a person changes it. That makes the network predictable, needs no extra processing or bandwidth for route-sharing, and avoids messages an attacker could fake. It suits small networks whose layout rarely changes. The weakness is that it cannot adapt: if a link fails, the router keeps sending packets down the dead path until someone notices and edits it, and in a big network editing every router by hand does not scale.

With dynamic routing the routers talk to each other using a routing protocol (RIP is one example), telling their neighbours what they are connected to and what it costs. Each router uses the news to build the cheapest routes (the idea is the same as finding the shortest path on a map), and when something changes, such as a link going down, the news spreads and every router recalculates automatically. It copes with growth and failures with no manual work. The costs are more processing, memory and bandwidth for the chatter, a more complex set-up that needs skilled staff, a delay while all the routers agree on the new picture (convergence), and the risk that the routing messages are attacked if they are not secured.