Component-based simulation

Moore's law states that transistor density doubles approximately every eighteen months. Abetted by an increase in clock speed, it contributes to an improvement at an even faster rate in microprocessor overall performance. Based on this hardware improvements, parallel simulations nowadays are theoretically capable of executing far larger and more complex models than ever before. However, there is a gap between the size of the simulation model that the state-of-the-art computers can execute and the size and complexity of the simulation model that the programmers can reliably design and build. Simulations are hard to program because programmers need to take the time semantics into consideration. Simulation programs are hard to debug because a procedure might be executed millions of times, each execution associated with different arguments and a distinct point in simulation time. More fundamentally, the two widely used simulation world views, event scheduling and process interaction, as discussed in this paper, don't scale well to large simulations. The modeling methodology has become an obstacle that limits the use of simulation technology outside research community. We believe that component-based simulation is a natural and intuitive approach to the development of large-scale simulations. In this paper, we first briefly review classical simulation world views and the logical process paradigm. After discussing their limitations, we present a component-oriented view that facilitates the use of the component-based approach in simulation development. To support this component-oriented view, we classify simulation components into timeless components, time-dependent components, and time-independent components. This classification promotes a hierarchical simulation modeling methodology that addresses both composability and efficiency.