End-to-end delay models for interactive services on a large-scale IP network
Michel Mandjes, K. van der Wal, Rob Kooij, H.J.M. Bastiaansen · 1999
In addition to the traditional best-effort Internet service, the Internet community now explores new Internet service models with guaranteed QoS levels. Work in this area has in particular been co-ordinated by the IETF working groups on Integrated Services (intserv) and on Differentiated Services (diffserv). The goal of these working groups is to define new QoS mechanisms for the Internet to be able to provide IP transport services with sufficient quality to support interactive services. A key aspect is the end-to-end delay that may be expected from using these services. Because no widely deployed networks exist which provide these low-delay services, this article addresses models for assessing the major components in the end-to-end delay. The models can be applied to both the intserv and the diffserv approach. An IP network configuration is considered consisting of a high-speed IP core network and two lower-speed access parts. For the core network queuing models are presented. In addition, it is concluded that, if some ‘loose requirements’ are fulfilled by the core network, the delay due to queuing can be kept below 8 ms; the propagation delay is to be added to this figure. For the access parts, very simple models can be used if it is assumed that only a single interactive application uses the access. Simple expressions are presented covering various protocols (including PPP with suspend/resume) to be used on the access parts.