Congestion control and service differentiation

Audio, video and multimedia services impose strict (real time) requirements on the loss, jitter and delay through the network. Some lesser requirements are set for services like web browsing, transactions and e-commerce. To meet these requirements, the routers use different traffic handling for the packets. For IntServ this is done on a per flow basis while for DiffServ it is done on a class basis. The latter therefore scales better and it is the preferred service architecture in the Internet. For Diffserv, the packet handling is divided into two sets of tools, the scheduling of packets, and the discarding of packets when buffers reach certain watermarks. Congestion control signals, either as dropped packets or through marking of packets, can be used to signal congestion to the sources and thereby let them adapt to the situation. This requires cooperation with the encoding at the source. Audio and video sources are studied for this purpose. Appropriate models of the sources have to be made, policies worked out for service differentiation in a router, and measurements taken.

In a wireless network, it is likely that re-routing either caused by mobility or link breaks will cause severe congestion. Adapting the source may then no longer be sufficient and connection may have to be closed down. In such situations, it is important that no new connections are accepted until the congestion is cleared. Otherwise the system will thrash, accepting new connections while closing down already established connections. The congestion control should therefore be coupled with a limited admission control.

Relations to external projects and activities:
EuroNGI, WP.JRA. 2.1 and WP.JRA. 2.2.

Project coordinator: Øivind Kure
Professors: Øivind Kure, Peder J. Emstad
PhD students: Erling Austreim, Bjørnar Libæk, Anne Nevin