Verteilte Constraint-basierte Eisenbahn-Simulation

Abstract

This work designs and evaluates a new algorithm for the simulation of railway systems: Distributed Railway Simulation (DRS). DRS is a distributed meta-al gorithm, that conducts local, constraint-based simulators, to cooperatively com pute a global simulation. For the analysis of complex real-world systems, often simulation approaches are used. Railway companies, like the German Deutsche Bahn AG, use simula tion software to examine and test their rail networks and timetables, in order to improve costs and service quality. This work took place within a ministry funded research and development project, where it should be investigated how Constraint Propagation can help overcome certain drawbacks of existing simula tion approaches. The long-term objective of the project was the rst microscopic simulation of the German rail network. For a distributed simulation with DRS, rst of all the simulation model has to be divided into a number of disjunctive parts. These parts are distributed among the available computing nodes. Then, DRS initiates on all these nodes a local simulation. After each node has completed its simulation computation, it compares its results with that of the neighboring nodes. As long as there exists an inconsistency between some neighbor's results, the local computations are repeated, now taking into account the results of the neighbors. This process is iterated until the global solution is consistent. This algorithm is described here formally in detail. Using the herein con structed formalism, I prove the two most important properties of the algorithm: termination (it distributedly computes a result after a nite number of steps, for all possible simulation con gurations), and correctness (each distributed com putation yields a correct simulation result). DRS has a well-founded theory, but has also been realized soundly. I derive requirements to the realization from state-of-the art literature. Based on the requirements, the system design is developed. The design is described using UML standards: architecture of the overall system, structure of the components, local and global sequences. Additionally, I go into the controller of the distributed computation (incl. formal proof of correctness of the implementation), into the integrated development process, and into the possibilities of using heterogeneous local simulation engines. The empirical investigations based on the available case study show that the algorithm and the implementation ful ll the aimed goals: The simulation of given railway systems takes less time when using DRS. This even applies to rather small nets and to only one computing node. But, the results even imply that DRS can simulate very large networks and therefore the German railway net in very short time: the system scales using a large number of computing nodes. In addition to the empirical results, the formal foundation is an outstanding property of this work. The presented approach can be used for a number of other applications and has propositions for some theoretical and practical extensions.