Distributed Agent-Based Traffic Simulations
| Autor: | Gerard Scemama, Omer Rana, Matthieu Mastio, Mahdi Zargayouna |
|---|---|
| Přispěvatelé: | Génie des Réseaux de Transport Terrestres et Informatique Avancée (IFSTTAR/COSYS/GRETTIA), Institut Français des Sciences et Technologies des Transports, de l'Aménagement et des Réseaux (IFSTTAR)-Communauté Université Paris-Est, School of Computer Sciences & Informatics [Cardiff], Cardiff University |
| Jazyk: | angličtina |
| Rok vydání: | 2018 |
| Předmět: |
Speedup
Computer science Distributed computing Computation DISTRIBUTED COMPUTING Context (language use) 02 engineering and technology SCALABILITY Modeling and simulation 0502 economics and business 0202 electrical engineering electronic engineering information engineering TRAFIC ROUTIER Fundamental diagram of traffic flow HIGH PERFORMANCE COMPUTING 050210 logistics & transportation Mechanical Engineering 05 social sciences Simulation modeling Supercomputer MODELISATION Computer Science Applications [INFO.INFO-MA]Computer Science [cs]/Multiagent Systems [cs.MA] Automotive Engineering Scalability SIMULATION SYSTEME MULTI-AGENT 020201 artificial intelligence & image processing |
| Zdroj: | IEEE Intelligent Transportation Systems Magazine IEEE Intelligent Transportation Systems Magazine, IEEE, 2018, 10 (1), pp. 145-156. ⟨10.1109/MITS.2017.2776162⟩ |
| ISSN: | 1939-1390 |
| DOI: | 10.1109/MITS.2017.2776162⟩ |
| Popis: | Modeling and simulation play an important role in transportation networks analysis. With the widespread of personalized real-time information sources, it is relevant for the simulation model to be individual-centered. The agent-based simulation is the most promising paradigm in this context. However, representing the movements of realistic numbers of travelers within reasonable execution times requires significant computational resources. It also requires relevant methods, architectures and algorithms that respect the characteristics of transportation networks. In this paper, we tackle the problem of using high-performance computing for agent-based traffic simulations. To do so, we define two generic agent-based simulation models, representing the existing sequential agent-based traffic simulations. The first model is macroscopic, in which travelers do not interact directly and use a fundamental diagram of traffic flow to continuously compute their speeds. The second model is microscopic, in which travelers interact with their neighbors to adapt their speeds to their surrounding environment. We define patterns to distribute these simulations in a high-performance environment. The first distributes agents equally between available computation units. The second pattern splits the environment over the different units. We finally propose a diffusive method to dynamically balance the load between units during execution. The results show that agent-based distribution is more efficient with macroscopic simulations, with a speedup of 6 compared to the sequential version, while environment-based distribution is more efficient with microscopic simulations, with a speedup of 14. Our diffusive load-balancing algorithm improves further the performance of the environment based approach by 150%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|