Program at a glance   Tutorial program   Technical program   Abstracts   Papers


Tracing and Measurement
Full Paper
Deriving Traffic Demands for Operational IP Networks: Methodology and Experience
Anja Feldmann (University of Saarbruecken)
Albert Greenberg (AT&T Research)
Carsten Lund (AT&T Research)
Nick Reingold (AT&T Research)
Jennifer Rexford (AT&T Research)
Fred True (AT&T Research)
Engineering a large IP backbone network is impossible without an accurate view of traffic demands. Shifts in user behavior, changes in routing policies, and failures of network elements can result in significant (and sudden) fluctuations in the flow of traffic through the backbone. In this paper, we first provide a model of traffic demands, carefully constructed for traffic engineering and performance debugging of large operational IP networks. In the Internet, a large fraction of the traffic is interdomain. We represent a demand as a volume of load from an ingress link to a set of egress links to capture how routing affects the flow of traffic between domains. Second, we provide a measurement methodology for computing traffic demands, combining flow-level measurements collected at all ingress links with reachability information about all egress links. Third, we extend our methodology to cope with practical limitations in operational networks on the amount of measurement data and the number of collection locations. Specifically, we show how to estimate traffic demands using measurements collected at a smaller number of edge links -- the peering links connecting to neighboring domains. For traffic entering the network on other links, we show how to infer the ingress link by combining the egress measurements at the peering links with a routing model that determines which path(s) the flow could have followed across the backbone. The traffic demand computations involve collecting, validating and joining a very large and diverse set of usage, configuration and routing data from multiple locations in the network, over the same time frame. We report on our experience in carrying out these computations. Finally, we analyze the dynamics of the traffic demands and discuss the implications on traffic engineering.