Computer Laboratory Home Page Search A-Z Directory Help
University of Cambridge Home Computer Laboratory
Thursday Nov 11th, 2004 - 4:30pm
Computer Laboratory > Research > Systems Research Group > NetOS > Seminars > Thursday Nov 11th, 2004 - 4:30pm

Random Trip Models

Milan Vojnovic
Random mobility models have been used extensively by a wide community of researchers for the purpose of analysis and simulation-based studies of mobile ad-hoc networks. It is thus surprising that some of the mobility models were ill-defined (non-existence of steady-state). Existence of a unique steady-state is important to evaluate long-run behaviour of a protocol under consideration. For well-defined mobility models, in general, starting from an initial distribution of the mobility state, there is initial transient phase during which the distribution of the state converges to a steady-state distribution. A common practice is to truncate the initial simulation run with the aim to eliminate the effect of the transient. The problem is that for some mobility models the initial transient lasts a long time -- for some models, as long as typical duration of a simulation run! The talk presents Random Trip Models, a broad class of mobility models that accommodates many existing mobility models in one; e.g. widely-used random waypoint. Random trip models are featured with having a unique steady-state distribution. We give a ready-to-use perfect sampling algorithm to sample the initial mobility state, so that mobility is in steady-state throughout a simulation, i.e. simulation is perfect. The algorithm alleviates knowing geometric normalization constants in the cases where they are difficult to compute -- a bound on diameter of the mobility domain suffices. Our perfect sampling is implemented in a tool to use with ns-2 network simulator. Joint work with Jean-Yves Le Boudec (EPFL), Santashil PalChaudhuri (Rice University) Reference: IEEE INFOCOM 2005 (to appear);