There are three
main approaches for identifying traffic to be switched:
Path creation can be control- or topology-driven, where labels are
preassigned in relation to normal routing control traffic. Here, the
network size dictates the load and bandwidth consumed by the
assignment and distribution of label information.
Request-based control traffic from protocols such as RSVP can
trigger path creation relating to individual flows or traffic
trunks. Here, the number of labels and computational overhead will
depend entirely on the number of flows being supported.
Data-traffic-driven label assignment is where the arrival of data
recognized as a flow activates label assignment and distribution on
the fly. This approach implies that there will be latency while path
setup takes place. Overheads in this case will be directly
proportional to traffic patterns.