Predictive control of delay and packet loss network QoS

1 Motivation and Introduction Real-time multimedia applications communicating over a packet-switched network present new challenges to the current networking technology. One of them is Quality of Service (QoS) requirement. Unlike some other network applications such as email and le transfer, which only require best-eeort service at the network layer, multimedia applications such as video-on-demand and multimedia virtual presentation 1] demand more strict QoS. Most multimedia applications require an explicit delay bound on their transmitted packets because of delay-sensitivity. For example, the delay bound of a live video transmission is related to the receiver's latest playback time: if the packets arrive beyond that point, they may be useless for the receiver. The delay bound also varies from one application to another due to the diierences in their delay-sensitivities and playback environment setups. While more and more multimedia applications such as vic 2], vat 3], nv 4], have been developed for a lossy packet-switched network environment such as that of the Internet, the packet loss tolerance of the application is also an important QoS issue: when network congestion happens, packets may be dropped due to an overrowing buuer or due to the eeort of bounding the delay. Those dropped packets directly aaect the viewing quality at the receiving end. Multimedia applications may have diierent packet loss tolerances due to the following two reasons: First, the applications may have diierent subjective quality requirements: a video conferencing session showing a person lecturing may tolerate more packet loss than a movie transmission session. Second, many schemes have been proposed to provide real-time network applications with better error recovery or protection over packet loss. Some of the approaches proposed for video transmissions include priority based protection and layered coding , structured packing, robust coding/decoding by introducing redundancy, and receiver recovery 5]. These schemes are diierent in their required conditions/equipments for deployment, their eeciency in ghting packet loss, and their bandwidth overhead and cost. A particular application may choose one or many of the above schemes to deal with packet loss. There is ongoing research to make some of these schemes more adaptive and conscious of the network environment and to improve their resilience to network packet loss 6, 7]. Because of the diversity of these schemes as well as the diierences in the subjective quality requirement of each application, it is likely that the multimedia applications will present diierent packet loss requirements to the network. In …