End-to-End Delay of Videoconferencing over Packet Switched Networks

Videoconferencing applications require from the network guaranteed bandwidth, small loss probability, and bounded delay. In order for the participants in a videoconference call to interact naturally, the end-to-end delay should be below human perception - about 100 ms. We identify the components of the end-to-end delay in various configurations with the objective of understanding how it can be controlled and reduced. The first contribution to the end-to-end delay comes from the processing performed on pictures before sending and after receiving them; we call this processing delay. The second contribution is the network delay. Since pictures must be displayed at the same pace at which they had been captured, any variation in the processing and network delay must be compensated before pictures are displayed. This compensation is done by adding resynchronization delay, which is the third component of the end-to-end delay. We devise these bounds, and hence the related end-to-end delay, going step-by-step through a number of configurations. We study the transmission of both raw video and MPEG video (VBR and CBR) over: (i) dedicated links, (ii) circuit switching, (iii) packet switching with time driven priority, and (iv) asynchronous packet switching. The study shows that a common time reference used with time driven priority can provide adequate results, for video conferencing with MPEG, independent of the network load.