Network Topology Inference with Partial Information

Abstract—Full knowledge of the routing topology of the In-ternet is useful for a multitude of network management tasks. However, the full topology is often not known and is instead estimated using topology inference algorithms. Many of these algorithms use Traceroute to probe paths and then use the collected information to infer the topology. We perform real experiments and show that in practice routers may severely disrupt the operation of Traceroute and cause it to only provide partial information. We propose iTop, an algorithm for inferring the network topology when only partial information is available. iTop constructs a virtual topology, which overestimates the number of network components, and then repeatedly merges links in this topology to resolve it towards the structure of the true network. We perform extensive simulations to compare iTop to state of the art inference algorithms. Results show that iTop significantly outperforms previous approaches and its inferred topologies are within 5 % of the original networks for all considered metrics. Additionally, we show that the topologies inferred by iTop significantly improve the performance of fault localization algorithms when compared to other approaches. Index Terms—Topology inference, Partial information, Fault localization. I