Increasing the coverage of a cooperative internet topology discovery algorithm

Abstract

Recently, Doubletree, a cooperative algorithm for large-scale topology discovery at the IP level, was introduced. Compared to classic probing systems, Doubletree discovers almost as many nodes and links while strongly reducing the quantity of probes sent. This paper examines the problem of the nodes and links missed by Doubletree. In particular, this paper's first contribution is to carefully describe properties of the nodes and links that Doubletree fails to discover. We explain incomplete coverage as a consequence of the way Doubletree models the network: a tree-like structure of routes. But routes do not strictly form trees, due to load balancing and routing changes. This paper's second contribution is the Windowed Doubletree algorithm, which increases Doubletree's coverage up to 16% without increasing its load. Compared to classic Doubletree, Windowed Doubletree does not start probing at a fixed hop distance from each monitor, but randomly picks a value from a range of possible values. © IFIP International Federation for Information Processing 2007.