10.5. Link Bundling
Link bundling has been proposed in the context of optical networks to improve the scalability of routing protocols. Neighboring nodes in an optical network are often connected by a large number of parallel links/fibers between them. With traditional IP routing protocols, each physical link between a pair of neighboring nodes constitute a routing adjacency. Hence, if there are N parallel links between two nodes, it will result in N routing adjacencies between them. Multiple parallel adjacencies severely limit scalability of routing protocols by creating unnecessary computational load on the network elements and superfluous control traffic.
In order to understand this problem better, let us consider the example in Figure 10-1, where two neighboring switching nodes in an optical network are connected by eight parallel links. Let us also assume that they are running the OSPF routing protocol. These eight parallel links between the nodes will result in eight routing adjacencies between them, with the OSPF Hello protocol running over each adjacency. Furthermore, anytime a node receives an LSA, it will send a copy to its neighbor over each of the eight parallel links. The neighboring node will accept only the first of these LSAs, discarding the rest. Clearly, this is a waste of network bandwidth and computing resources on the nodes.
Figure 10-1. Link Bundling
Link bundling eliminates this inefficiency by aggregating several (or all) of the parallel links between two neighboring nodes into a single logical link. The resulting logical link is called a bundled link (also known as a traffic engineering (TE) link in GMPLS OSPF/IS-IS terminology) [Kompella02a, Kompella02b], and the constituent physical links are called component links. This aggregation improves routing scalability by reducing the amount of routing information exchanged between neighboring nodes. As with any aggregation technique, link bundling may result in loss of information. In order to limit this loss, some restrictions on link bundling have been put in place.
For example, consider the scenario in Figure 10-1. The two switches in the figure are connected by two DWDM systems, each carrying four wavelengths. Note that the DWDM systems are connected by fibers that pass through two different conduits. Now, it is possible to bundle all eight links (wavelengths) between the two switches into a single bundle as shown (bundle 3). This bundling, however, fails to capture the fact that the links go over two different DWDM systems and through two different conduits, and hence form two different risk groups. A better approach would be to create two different bundles, bundle 1 and bundle 2, each consisting of links that use the same DWDM system and conduit.
All component links in a bundle must begin and end on the same pair of nodes. All component links must have the same link characteristics, for example, the same TE metric (i.e., an administrative cost). Although not mandatory, they should also belong to the same set of SRLGs. Aggregation rules can be defined for other parameters, such as the maximum bandwidth available on the link, the maximum bandwidth that can be allocated to a connection traversing the link, and so on. Bundling may also be applied recursively, that is, a component link may itself be a bundled link.