On the Utility of Inference Mechanisms

A number of network path delay, loss, or bandwidth inference mechanisms have been proposed over the past decade. Concurrently, several network measurement services have been deployed over the Internet and intranets. We consider inference mechanisms that use O(n) end-to-end measurements to predict the O(n²) end-to-end pairwise measurements among n nodes, and investigate when it is beneficial to use them in measurement services. In particular, we address the following questions: (1) For which measurement request patterns would using an inference mechanism be advantageous? (2) How does a measurement service determine the set of hosts that should utilize inference mechanisms, as opposed to those that are better served using direct end-to-end measurements? (3) How can the answer to question 2 be efficiently computed as measurement requests arrive and terminate? Our solution is able to identify groups of hosts which are likely to benefit from inference, by utilizing a probabilistically generated spanning forest on the measurement request graph. We compare our solution to a simple heuristic that uses the number of measurements a host participates in. Results with synthetic datasets as well as datasets from a popular peer-to-peer system demonstrate that our technique identifies host subsets that benefit from inference quite accurately, and in significantly less time than an algorithm that identifies optimal subsets. The measurement savings are large when measurement request patterns exhibit small-world characteristics, which is often the case for peer-to-peer and other popular distributed systems.