A CORBA-Based Distributed Simulation Methodology for Hierarchical DEVS Models

Discrete-event simulation is frequently used to analyze and predict the performance of systems. Simulation of large, complex systems remains a major stumbling block, however, due to the prohibitive computation costs. Distributed simulation offers one approach that can significantly reduce these computation costs. Since distributed simulation deals with large and complex systems, the following issues should be addressed: model verification and validation, model reusability, and user-transparency of distributed simulation details, etc. The Discrete Event Systems Specification (DEVS) formalism [1,2,9] developed by Zeigler is a formal framework for specifying discrete event systems. The DEVS-based distributed simulation approach [4-8,12-15] can be an attractive alternative to the conventional distributed simulation approaches by its independence of any computer language implementation and its modular, hierarchical modeling methodology. This paper presents a new distributed simulation methodology for models specified by the DEVS formalism. The so-called hierarchical simulation mechanism has been the only simulation mechanism for DEVS models. In this mechanism, scheduling of models and routing of input/output events are done hierarchically. Most of all Distributed simulation approaches for DEVS models have been based on this mechanism. In respect of the synchronization in distributed simulation, the Time Warp mechanism [10-11] has mainly used in the approaches, thereby forming the hierarchical Time Warp, which is more complex than the conventional Time Warp. In this paper, we first propose a non-hierarchical simulation mechanism for DEVS models. It transforms hierarchically structured DEVS models into non-hierarchical ones. There is a central scheduler that schedules the entire DEVS models. Input/output event routing is done directly by the sending model itself. This transformation can eliminate the overheads incurred during the conventional hierarchical simulation mechanism. Based on this proposed non-hierarchical simulation mechanism, we then propose a distributed simulation environment for DEVS models, called DEVSCluster. DEVSCluster employs the Time Warp mechanism for the synchronization and the non-hierarchical simulation mechanism for scheduling of DEVS models. Compared to the previous distributed DEVS simulation approaches, DEVSCluster is very stable due to its relatively simple scheduling and synchronization capabilities. Another important characteristics of DEVSCluster is the employment of CORBA-based input/output routing. Object Management Group’s (OMG’s) Common Object Request Broker Architecture (CORBA) provides location-transparent distributed object activation and messaging. The use of CORBA implies that DEVSCluster is basically an object-oriented environment. Simulation objects can communicate with one another by a simple method-call (That is, not an explicit exchange of real message), regardless of whether they reside in the same node or not. This can speed up the simulation since the method-call can reduce significantly the overhead of making a real message even in the intra-node communication. To show the effectiveness of the proposed methodology, we realize DEVSCluster in Visual C++, and conduct a benchmark simulation for a large-scale logistics system. In the logistics system, wares, stores, and control centers are located over wide areas: stores spend products, wares supply the products, and control centers control the logistics cars to chain the stores and wares. In this system, the objective is to minimize the number of the required cars while satisfying the orders of all stores. We compare DEVSCluster and the previous DEVS simulation approaches. The configuration of the simulation platform is 4 Pentium III-based Windows NT systems connected by the 10Mbps Ethernet. We implemented the simulation system in Visual C++ 6.0 and the GUI by JAVA. The performance result shows that the proposed methodology works correctly and performs better than the previous approaches.