Matnet Matlab Network Simulator for Tiny OS Alec

Matnet – Matlab Network Simulator for Tiny. OS Alec Woo Terence Tong July 31 st, 2002

Motivation ¨ Need a simulation environment to quickly analyze and compare protocols – Tossim: best for debugging TOS code – NS: missing the integrated analysis tools ¨ Prowler: – TOS network simulator in Matlab • by a grad student, Gyula Simon, at Vanderbilt University – A solution? (No…) • Basic framework can be more decoupled • Easier to borrow ideas but start from scratch

Overview ¨ Mat. Net provides – Packet level simulation • With probabilistic loss and collision model based on empirical data – TOS MAC layer implementation • With fixed or infinite transmission queue – Event driven programming like TOS – GUI support • Real time visualization of topology evolution – Interactive environment • Pause and resume for debugging

Overview (Cont. ) ¨ Layering architecture allows plug and play of different modules • radio model (provides generic_comm interface) • routing protocols (provides routing services) • application (user application) ¨ Standard analysis utilities • Hop distribution, path reliability distribution, bandwidth distribution, aggregate bandwidth ¨ Support for periodic snapshot of the network – Can easily generate movies ¨ Topology generation – Grid or random or file input

Initial Results ¨ Protocols simulated: • Broadcast based routing • Surge like shortest path routing • Maximum reliable path routing ¨ Application simulated: • Collect data ¨ 49 -node on a gird with ~15000 simulated messages takes about (P 4 1. 8 GHz) • 9 minutes in Matlab • 7 minutes with Matlab->C in cygwin • 6 minutes with Matlab ->C in Windows command

Global behavior observation ¨ Broadcast based flooding – resembles empirical result done last year • backward links • long links ¨ Other distributed routing algorithms: – surge shortest path – maximum reliable path routing • resemble results from static analysis

Empirical Data

Model at Medium Power clear transition silent

Connectivity Contour from Model