On the Cost of Network Inference Mechanisms
An important class of network inference mechanisms estimate the properties (e.g., delay or loss) of a large number of end-to-end network paths by measuring some subset thereof. This class of mechanisms is designed to reduce the amount of injected active measurement probe traffic and effort required to collect a large set of measurements, usually at the expense of measurement accuracy. For example, the Azureus BitTorrent client can use inferred network delay information to select peers from which to transfer data. Another example application is the UUSee television streaming service.