Provisioning Internet backbone networks to support latency sensitive applications

Interactive applications such as voice, audio, and video, as well as business applications such as Virtual Private Networks are becoming an increasingly important component of Internet traffic. Such applications have strict requirements on the total end-to-end delay which may be incurred in the network. One approach to meeting these delay requirements, known as service differentiation, is to give preferential treatment to latency sensitive traffic. Doing so may lead to an efficient network design with the minimum amount of resources (e.g., bandwidth) required to support the needs of each traffic type. An alternative approach is to provide sufficient resources so that all traffic meets the most stringent delay requirements. This latter approach is known as overprovisioning.; In the context of wide-area Internet backbone networks, two factors make overprovisioning an attractive approach. First, the high link speeds and large volumes of traffic make service differentiation complex and potentially costly to deploy. Second, given the degree of aggregation and resulting traffic characteristics, the amount of overprovisioning required may not be very large. We establish that this is indeed the case by collecting and analyzing traffic measurements from the Sprint IP network, a commercial Tier-1 Internet backbone.; We begin by performing network simulations using a set of 331 one-hour traffic measurements from the Sprint network. These simulations demonstrate that link utilization can reach 80%–90% before queuing delays begin to exceed several milliseconds. While the simulations can be used to evaluate the level of overprovisioning required in the network, many network design problems are greatly aided by an analytic traffic model. We therefore develop a traffic model which captures the observed characteristics of backbone traffic and derive expressions for the delay through a single queue in the network, as well as the end-to-end queuing delay. Using this model we solve several network design problems including capacity planning and optimal route selection.