Discovering network topology by correlating end-to-end delay measurements

Abstract

Conventional methods of discovering network topology require the cooperation of network elements. We present a method of network topology discovery based solely upon end-to-end delay measurements that requires only the cooperation of end systems. Previous work using end-to-end measurements has focused on discovering tree topologies; the method here discovers more general networks. The discovery method is based on two algorithms: the matroid algorithm and the correlation algorithm. This work develops and validates the correlation algorithm for three increasingly sophisticated network models using simulation. We also develop an indicator of the quality of a particular result.