A distributed modelling framework for large-scale microscopic traffic simulation. (20th February 2009)

• Record Type: Journal Article
• Authors: Liu, Henry X.; Ma, Wenteng
• Is Part Of: World review of intermodal transportation research. Volume 2:Number 2/3(2009)
• Page Start: 107
• Page End: 126
• Publication Date: 2009-02-20
• Subjects: microscopic traffic simulation -- distributed processing -- large-scale networks -- API -- application programming interface -- distributed modelling; Containerization -- Research -- Periodicals; Transit, International -- Research -- Periodicals; Freight and freightage -- Management -- Periodicals; 387.5442068
• Languages: English
• ISSNs: 1749-4729

Abstract

Simulation modelling is an increasingly popular and effective tool for analysing transportation problems, which are not amendable to study by other means. In this paper, we examine the need for distributed simulation from the perspective of computational speedup and development of large-scale network simulation contexts and datasets. After an overview of the general techniques for the distributed discrete-event simulation and previous efforts on the distributed traffic simulation, we present the general architecture of the proposed distributed modelling framework. Two categories of modelling strategies, namely, light global control/independent subnets versus heavy global control/coordinated subnets are described. We have implemented the distributed scheme of light global control/independent subnets and the implemented details, such as communication techniques and vehicle transferring across the boundary of two subnets are discussed. Unlike the previous studies using the dedicated high performance machines, our efforts are to utilise the low-cost networked PCs that are commonly available. By using the Application Programming Interface (API) functions supported by off-the-shelf microscopic traffic simulation software Paramics, we are able to distribute the computational load of microscopic simulation to multiple single-processor PCs without access the proprietary source codes of the simulation program. Performance testing and analysis of the implemented prototype demonstrate that the proposed framework is very promising.