On the aggregation of deterministic service flows

It is common belief that "flat" Quality of Service (QoS) architectures, e.g. the IETFs Integrated Services architecture (IntServ), are not scalable to large networks, e.g. the global Internet. This is due to the ambitious goal of providing per-flow QoS and the resulting complexity of fine-grained traffic management. One solution to this problem is the aggregation of traffic flows in the core of the network, thus creating a hierarchical resource allocation system. While one might suspect that aggregation leads to allocating more resources for the aggregated flow than for the sum of the separated flows if flow isolation shall be guaranteed deterministically, we show in this article that for IntServ's Guaranteed Service flows this is not necessarily the case even if flow isolation is retained. We compare different approaches to describe the aggregated traffic and analyze their impact on bandwidth consumption and ease of flow management. Furthermore, we perform a thorough numerical evaluation of the derived results with respect to their behavior in response to changes in exogenous parameters like the traffic specifications of the flows or the configuration of the network. Applications of these theoretical insights and numerical evidence could be to use the derived formulae for resource allocation in either a hierarchical IntServ, IntServ over DiffServ (Differentiated Services), or IntServ over ATM (Asynchronous Transfer Mode) network