Routers, unlike bridges, connect networks into an internetwork in which each network retains its logical identity. It is similar to the way a freeway will connect far away cities which are still distinct. We wouldn't say that San Diego is the same as Los Angeles, just because Interstate 5 runs through them both.

An internetwork based on routing consists of many different logical subnetworks, each of which is a potentially independent administrative domain. Today, router generally means multiprotocol router - that is, a router that can handle multiple protocols. A router must have appropriate software for each protocol it supports because, unlike bridges, routers are active devices. This means that they may make several decisions about each packet such as whether to discard it because it has taken too long to reach its destination. For this reason, routers need to know more about protocols than bridges.

In discussing routers, it's important to know the Open Systems Interconnection (OSI) reference model. The OSI model describes an internetwork as consisting of seven (7) layers, from logical to physical. Level 1 is physical, level 2 is datalink (such as bridges), and level 3 is the network layer. Routers operate at this network layer, and are concerned not only with the source and destination address of each packet, but also with the actual path that each packet takes through the network. Routers have more decisions to make than bridges, and accordingly they need more information such as the cost of transmitting the packet. This information is contained in the router's database, known as a routing table. This database is different from the address database found in bridges, because the routing table includes information on the paths, or routes, that any packet can take through the network to get from its source to its destination.

The basic functionality of a router is to create and maintain the routing table, and secondly, to select the next course of the journey for each packet, based on the information contained in the packet and the routing table. Routing tables can be created either statically or dynamically. With static routing, the network administrator must manually build the routing table by making a database entry for each segment of each possible path through the network. With dynamic routing, the routing table is automatically constructed by the router by using special packets containing path-oriented information. A router may also issue packets containing updates when it detects a change in the network.

Although static routing may be advantageous in environments that require absolute security, dynamic routing is more common. To help create and maintain routing tables, a dynamic router broadcasts information when it detects a change in the network. Such information might specify the existence of a new path through the network or the removal of a path from service. Both the amount of routing information and the number of routers to which it must be sent depend in part on the protocol involved and what type of routing algorithm it uses.

When a router receives a packet, it examines the destination of the packet and determines which path is the best one to send it along. This determination depends on several factors. 1) The measure of distance, or "routing metric" in use. 2) The core algorithm implemented by the high-level protocol being used. 3) The architecture of the routed network.

In terms of distance only, the best path through the network is the shortest path. This seems obvious enough until one considers how to define distance. In most cases, the purely geographical distance between any two points is not a very good way to compute a path through a network. It does not typically correspond to the way the information is actually sent. Nor does it take into account other important factors such as economic costs. Instead, routers more commonly use the "number-of-hops" routing metric to calculate the best path. Using this metric, a router computes the best path through the network based on the number of router-to-router transmissions, or hops, each path requires. The best path is defined as the one requiring the smallest number of hops. However, since this metric does not factor in other variables such as cost, line speed, and transmission delay, more complete routing metrics have been developed to include these variables.

Routers impose no constraints on network topologies. They can be used to implement all the topologies described in last month's article on bridges. More importantly, they can utilize active paths. Redundant paths can be used to split the traffic load. Active loops pose no problems for routers because they will always determine appropriate paths.

In the event that traffic congestion does occur, routers can handle the problem in several different ways. Since routers analyze the contents of a packet's control fields and make routing decisions accordingly, they can also change the contents of these fields. For example, a packet contains information called a "time to live" field which is modified by each router as it forwards the packet. If the packet has been around too long the router will automatically discard it and wait for the source systems to regenerate it at a later time. Another solution to the problem is to use a "source quenching" technique allowed by some protocols. This technique allows routers to temporarily stop packets originating from any particular source until the congestion is relieved.

Because routers logically isolate internetwork segments, they allow the implementation of "firewalls." Firewalls are logical barriers that isolate segments and protect the rest of the network from incidents like conflicting applications or periods of traffic congestion that occur within a particular segment. Router-based internetworks contribute significantly to fault isolation and error recovery characteristics of good network design, which in turn enhances network security.