LINK INFERENCE IN LARGE NETWORKS BASED ON INCOMPLETE DATA

A network is partitioned into a set of independent partitions, and the topology of each partition is determined, then merged to form a topology of the entire network. Preferably, the partitioning is hierarchical, wherein the network is partitioned to form individual VLAN partitions, and each of the VLAN partitions is further partitioned based on the nodes that are simply connected to each port of one or more selected root switches within the VLAN partition. Simple connections to each port are efficiently determined based on an aggregate address forwarding table associated with each node. Ancillary information, such as spanning tree or CDP data, may be used to facilitate efficient partitioning and/or to validate inferences that are made with incomplete information.