Routing in Wide Area Networks

1. For networks that are very large, either in terms of the number of hosts or the distances between hosts, the "shared medium" approaches used in LANs just don't scale.

   • No matter how efficient the protocol for sharing the medium may be, having all message go to all stations involves too much unnecessary transmission of information when the number of stations is very large.

   • We have seen that the efficiency of most medium sharing protocols decreases as the physical size of the network increases.

2. In such networks, it is better to depend on intermediate nodes to forward packets selectively along paths leading to their destinations. This is similar to the work performed by bridges on a LAN except:

   • We are no longer concerned that the process be transparent to sending stations or to network administrators, so...

   • We would rather use multiple (i.e. cyclic) paths in the network to maximize performance and improve reliability rather than depend on a spanning tree.

3. To make all this work, when machine A wants to talk to machine B, somehow the network has to find a path from A to B. There are several major characteristics of the process of finding such paths that vary in significant ways from network to network:

   Dynamic Nature of Routing

   The book talks of the choice between establishing static routes when a network is created or using algorithms that dynamically vary routes as components fail, become busy or even get repaired. In fact, I think it is best to say that static routing is something of a strawman whose main role should be to point out the need for dynamic routing. In a large network, the ability to adapt to changes in network load and component failure makes routing protocols that adjust to changes in the state of network links essential.

   Distributed Routing

   In some early networks a central system computed routes for the whole network. Such a system is both a threat to the reliability of the network and a bottleneck at high load. The alternative is to somehow distribute the task of computing routes among all the nodes of the network. While this is preferable to the alternative of a centralized system, we will see that much of the complexity of the routing problem is due to the use of distributed algorithms.