Abstract: Simulation is a cornerstone technology for studying Vehicular Ad Hoc Networks (VANETs), yet the choice of mobility model can significantly alter simulation outcomes. This paper compares the simplistic Random Direction Model (RDM) with the realistic Krauss car-following model within a simulation system built on SUMO and NS2. By evaluating the AODV routing protocol, the study demonstrates that the Krauss model provides superior performance metrics, highlighting the critical impact of realistic mobility modeling on network protocol validation.
Introduction
Designing reliable communication protocols for VANETs is inherently difficult due to high vehicle speeds and restrictive road topologies. Researchers rely on simulations to test these protocols, but the validity of these tests depends on the underlying mobility model. If a model assumes vehicles move randomly without regard for roads or other traffic, the resulting data fails to reflect real-world driving conditions. This study addresses the gap between oversimplified "random" models and complex "car-following" models to determine how much the choice of model skews the evaluation of the Ad Hoc On-Demand Distance Vector (AODV) routing protocol.
Methodology
The authors developed a multi-stage simulation pipeline to bridge traffic and network dynamics:
- Map Generation: Real-world geographic data for a 1.4km x 2.0km area of Hohhot city, China, was extracted from OpenStreetMap (OSM) and refined using JOSM.
- Traffic Simulation: Using SUMO (Simulation of Urban MObility), vehicle movements were generated based on two models:
- Random Direction Model (RDM): Vehicles move to random destinations independently of roads or other vehicles.
- Krauss Model: A microscopic car-following model where speed is determined by the preceding vehicle's behavior, accounting for safe braking distances and driver proficiency.
- Network Simulation: Movement traces were imported into NS2, where the AODV protocol managed data transmission across varying node densities (40 to 120 nodes).
Key Insights and Analysis
The study reveals several critical differences in how these models affect network topology:
- Topology and Stability: In RDM, nodes are scattered across the map. In the Krauss model, nodes are constrained to roads. This concentration actually stabilizes the network, as nodes remain within communication range more consistently.
- Temporal Dependency: The Krauss model acknowledges that a vehicle's state depends on its previous state and its neighbors. This "physical common sense" leads to more predictable and durable communication links.
- Density Dynamics: A transition point was observed around 60–80 nodes. Below this, both models suffer from link breakages due to low density. Above this, the clustering inherent in the Krauss model allows for significantly more efficient multi-hop routing compared to the scattered RDM approach.
Results
The Krauss model consistently outperformed RDM across all primary performance metrics:
- Average End-to-End Delay: Krauss exhibited lower latency. For instance, at 60 nodes, Krauss had a delay of ~0.18s compared to RDM’s ~0.25s.
- Packet Loss Ratio: While loss dropped as density increased for both models, Krauss maintained a lower rate because its car-following nature kept nodes in more reliable, proximal chains.
- Throughput: Krauss achieved higher throughput, reaching nearly 150,000 B/s at 120 nodes, whereas RDM lagged behind.
Conclusion and Future Work
The experimental results confirm that the Krauss model is more effective for validating AODV routing protocols in realistic urban environments. However, the authors note that even the Krauss model is a simplification. Future research should explore more complex models like the Intelligent Driver Model (IDM), incorporate environmental obstacles (e.g., buildings blocking signals), and test a wider variety of protocols, such as Geographic Routing (GPSR), to further refine VANET communication strategies.
Summary updated