<title>Impact of traffic handling on Internet capacity</title>

Abstract

This paper describes the impact of traffic handling mechanisms on network capacity for support of Quality of Service (QoS) in the Internet. The emergence of applications with diverse throughput, loss and delay requirements requires a network that is capable of supporting different levels of service as opposed to the single best-effort service that was the foundation of the Internet. As a result the Integrated Services (Intserv) and Differentiated Services (Diffserv) models have been proposed. The Intserv model requires resource reservation on a per-flow basis. The Diffserv model requires no explicit reservation of bandwidth for individual flows and instead relies on a set of pre-defined service types to provide QoS to applications. Flows are grouped into aggregates having the same QoS requirements and the aggregates are handled by the network as a single entity with no flow differentiation. We refer to this type of handling as semi-aggregate. The Best-Effort model does not perform any differentiation and handles all traffic as a single aggregate. Each of these traffic handling models can be used to meet service guarantees of different traffic types, the major difference being in the quantity of network resources that must be provided in each case. In this paper, we consider the issue of finding the cross-over point at which the three approaches of aggregate traffic management, semi-aggregate traffic management and per-flow traffic management become equivalent. Specifically, we determine the network capacity required to achieve equivalent levels of performance under these three traffic management approaches. We use maximum end-to-end delay as the QoS metric and obtain analytic expressions for network capacity based on deterministic network analysis. One key result of this work is that on the basis of capacity requirements, there is no significant difference between semi-aggregate traffic handling and per-flow traffic handling. However Best-Effort handling requires capacity that is several orders of magnitude greater than per-flow handling.