Service differentiation

The existing Internet, based on the Best Effort principle (see [ H 2 285 ]), was not initially designed to take into account Quality of Service (QoS) information. The Best Effort model maximizes resource utilization while simplifying the operation of interconnection equipment. When a router's memory is saturated, incoming packets are discarded. TCP (Transport Control Protocol) retransmission mechanisms recover these losses, guaranteeing fault-free transfer from end to end. At the same time, TCP's "slow start" and "congestion avoidance" algorithms ensure a relatively equitable distribution of resources .

While TCP's behavior meets the needs of traditionally network-intensive applications such as telnet or ftp , it is poorly suited to the transmission of time-constrained flows. Most videoconferencing or Voice over IP (VoIP) applications are based on the User Datagram Protocol (UDP) . This protocol can be used to achieve near-constant throughput, but it does not implement a congestion management mechanism. The presence of non-adaptive flows affects TCP performance, penalizing applications based on this protocol. Control measures are needed to avoid network saturation by non-adaptive flows, ensuring that a minimum of resources is available for TCP applications to communicate.

TCP's control mechanisms prevent resource sharing. The amount of resources allocated to a flow depends on link capacity and the presence of other flows in the network. Economic interests are pushing for a change in this behavior. A controlled distribution of bandwidth would enable operators to define new services...