QoS information approximation for aggregated networks

Many important network functions (e.g., QoS provision, admission control, traffic engineering, resource management) rely on the availability and the accuracy of the network state information. However, it is impractical to maintain the complete state information of a large internetwork at a single location. Instead, a large network is often hierarchically structured, with each domain advertising its aggregated state. To achieve scalability, the amount of information after aggregation should he minimized. To improve accuracy, the aggregation method must be carefully selected. This paper gives a unified account of state aggregation based on the concept of service curves. The aggregation of network state is modeled as a recursive process of service curve transformation. New approximation methods based on polynomial curves, cubic splines and polylines are proposed, and their scalability/accuracy tradeoffs are studied. Our simulations show that these new methods approximate the network state far more accurate than the existing methods. In particular, the polylines achieve the best scalability/accuracy tradeoff.