An agent-based data collection architecture for distributed simulations

This paper presents an agent-based architecture for data collection and output analysis within a distributed simulation. Typically, a distributed model is decomposed into submodels that exchange entities such as system states, transitions, state durations, and objects. These submodels may run on different computers connected through a network. Because each submodel operates in a relatively independent manner, they each have only a partial view of system characteristics. Hence it becomes necessary to provide services that facilitate data collection and analysis within the distributed simulation. The main contribution of this article is to enhance the software support to enable data collection and analysis within the distributed simulation. The framework proposed in this work is integrated within the DEVSJAVA environment, which provides a system-theoretic basis for formulating discrete event simulation models through a collection of objects. The proposed system provides a scalable distributed computing architecture. The operation of the system is illustrated through an example in automated mining. We compare our architecture with a baseline methodology in which all submodels report all data to a central database for output analysis. The experimental results show that the proposed architecture can reduce network traffic significantly, provided the data collection agent is able to choose its operational parameters appropriately. The results also show that the system is able to work well with both heavy and light network traffic cases.