Network Capacity Trade-offs for Traffic Aggregation in Future Networks
Project Award Date: 05-01-2000
Evaluation of traffic handling mechanisms for support of Quality of Service (QOS) on the Internet will be a continuing technical issue, impacting network engineering and design. Identifying the tradeoffs associated with the use of traffic handling mechanisms, with respect to network capacity, as router and optical technology continue to advance, will be of value. By incorporating sensitivity analysis, this research will provide a tool for long-term planning to show how the various traffic-handling mechanisms will react to growth in network traffic.
Traffic handling mechanisms can be broadly classified as aggregate, semi- aggregate or per-flow (zero-aggregation). In aggregate traffic handling, there is no differentiation between traffic flows, and resources are allocated to the entire set of flows as a whole. With semi-aggregate traffic handling, traffic is grouped into a small number of predefined classes based on certain criteria such as the nature of delay guarantees required by the traffic. Resources are then allocated to each class of traffic. With per-flow handling, there is no grouping of traffic, and each flow is allocated its own dedicated resources. The choice of which traffic-handling strategy to use requires a methodology that can be used to capture the trade-off between the different schemes; to develop such a methodology is the purpose of this study. ITTC has developed an initial methodology to address these trade-offs and has conducted a preliminary analysis, which has demonstrated that it is possible to quantify the trade-off between traffic management and network capacity.
There are several ways this project is applying and extending the analysis to fully address the traffic-management-complexity-versus-network-capacity tradeoff. The tasks are as follows:
- (1) Review the methodology, used in the proof of concept study, which is based on a single-node network. Adjust the methodology to cater for network topologies of arbitrary size and numerous flows.
(2) Design a network topology that is representative of carrier networks, in terms of the number of nodes involved, number of hosts attached, and profile of applications supported.
(3) Look at the application of Network Calculus to carrier-sized networks of arbitrary topology. A significant aspect of this task will be to determine how end-to-end delays can be allocated to different layers in a hierarchical network.
(4) Address each scheme's sensitivity to changes in network traffic.
(5) Examine existing models that use stochastic bounds in the description of the user traffic; explore modifications to existing models. Determine to what extent the use of stochastic bounds on the traffic models affects the difference in capacity requirements of the traffic handling mechanisms.
Faculty Investigator(s): Victor Frost (PI)
Primary Sponsor(s): Sprint Corp.