Motion Pattern Characterization NSF Wireless Mobility Workshop Rutgers, July 31-Aug 1, 2007 Mario Gerla Computer Science Dept, UCLA
Why Motion Characterization? Different protocols depend on different motion characteristics –Predecessor based routing (eg, AODV, etc) depends on “link” lifetime –Georouting depends on neighborhood density and stability –Epidemic dissemination benefits from rapidly changing neighborhood Ideally, we would like to compare experiments run in different cities/scenarios -It would be nice to define a mobility “invariant” that guarantees consistency across different scenarios
Case Study: Epidemic Dissemination of data sensed by vehicles Designated Cars (eg, busses, taxicabs, UPS, police agents, etc) –Continuously collect images on the street (store data locally) –Process the data and detect an event –Classify the event as Meta-data (Type, Option, Location, Vehicle ID) –Epidemically disseminate (ie distributed index implementation) –Agents harvest the field Meta-data : Img, -. (10,10), V10
Epidemic Experiments (via Simulation) Simulation Setup –NS-2 simulator –802.11: 11Mbps, 250m tx range –Average speed: 10 m/s –Mobility Models Random waypoint (RWP) Real-track model (RT) : –Group mobility model –Probabilistic merge and split at intersections Westwood map
Mobility Models Track ModelRandom Waypoint Model
Meta-data harvesting delay with RWP Higher speed improves dissemination and reduces harvest latency Time (seconds) Number of Harvested Summaries V=25m/s V=5m/s
Harvesting Results with “Real Track” Coordinated motion patter slows down dissemination, increasing latency Time (seconds) Number of Harvested Summaries V=25m/s V=5m/s
Data Dissemination Efficiency The data dissemination efficiency depends on: –The rate by which a vehicle encounters neighbors proportional to velocity and density –The fraction of vehicles that are new Dependent of motion pattern and grid topology Can we define a single universal metric that captures motion patter and topology ? Enter: Neighborhood Changing Rate (NCR)
Neighborhood Changing Rate (NCR) Let’s define – : Sampling interval equal to the time needed for a node to move a distance equal to its transmission range – : Neighbors that entered node i’s neighborhood at the end time interval – : Neighbor that have left node i’s neighborhood at the end of time interval – : Node i’s nodal degree at time t. Then,
Manhattan one-way grid NCR varies from 0 to 1 depending on the routing at the intersections
Neighborhood Changing Rate (NCR) NCR depends only on Topology and Mobility Patterns Given average speed, density, and NCR, we can –perform cross-topology and cross-mobility patterns performance evaluations/comparisons –Predict efficiency of epidemic dissemination in said scenario
Experimental “validation” of NCR Data Dissemination ProtocolMobeyes SimulatorNS-2.27 Hello Intervals3[s] Data Generation intervals10’000[s] Simulation time2000[s] Simulation Area2400[m] x 2400[m] Number of Nodes100 Tx Range250[m] Speed5[m/s], 15[m/s], 25[m/s] Simulation Parameters Simulation Environment Urban Map TopologyTriangle Topology 760m 2400m
More Mobility Models… Track ModelRandom Waypoint Model [1] Biao Zhou et al. University of California, MilCom 2004
Harvesting Efficiency vs NCR NCR on a Map Topology with a speed of 5 m/s
Latency: different scenarios but same NCR Latency for scenarios with same speed, density and NCR, and for different mobility models and topologies
Conclusions and Future Work NCR can help compare/predict epidemic performance Future uses of NCR: –P2P Propagation of NCR, density and velocity parameters in the urban grid –Estimation of epidemic latency; does it make sense to disseminate? Can we define NCR-like invariants for other protocols/applications?