Label switching and routing

A key idea in label switching is that the use of labels can effect some centralization in route determination. Unlike IP-only routing algorithms, which can only make single-hop determinations, a label switched environment can create multiple-hop path segments or tree structures by distributing appropriate label information. Typically, labels are generated for a set of covering paths or trees for some network, and incoming packets are labeled for the appropriate path or tree. The path determination then needs only to be done once or some limited number of times; at the intermediate nodes, the label alone determines the next hop.

This has a number of advantages:

• Single-hop route determinations are subject to various anomalies such as routing loops which can be difficult to detect and fix. With labels, we can generate and maintain individual routes and route segments as desired, forestalling the development of any loops or transient misdirection of packets.

• When separate routing decisions are being made at every node, it is very difficult to ensure that end-to-end quality of service constraints are met. When labels are used, we can, through control of label generation and assignments, coordinate the necessary resource reservations and queue prioritizations to guarantee bandwidth levels and control overall packet delay times.

• Also, since routing decisions can be made at fewer points, and even made a priori (in whole or in part), there is more room for implementing more complex routing algorithms which may be capable of providing better path determinations and better overall network utilization. In particular, the idea of routing via a collection of moderate-length path segments or trees is an area as yet relatively unexplored even theoretically; until now, most routing algorithms have assumed either single-hop or end-to-end decision-making.