The second stage is the transmission itself. This delay can be predictable for circuit-switched networks, such as ISDN, or unpredictable in packet-switched networks. For example, in IP networks, packets are delivered in a store-and-forward way using buffering, and whenever congestion occurs, routers can discard packets. It is thus not a rare occasion when packets corresponding to the same video bitstream are following different routes, experiencing different delays, and arriving out-of-sequence at the receiving end.
Last, delay is introduced during decoding, as there is usually buffering of the reference frames for motion compensation. This delay has been kept to non-critical levels, through the reordering of frames before transmission, and the availability of cheap computing power.
In two-way videoconferencing systems, the total, or transit delay is critical. A delay of more than a few hundred milliseconds makes conversations and interactivity frustrating for the communicating parts. In one-way communication systems, however, the delay variation is more critical, as it has an increasing effect to the buffering requirements at the decoder, and consequently its cost.