Metamodel driven acceleration of actor-based simulation

Agent-Based Modeling (ABM) is a simulation technique that captures model interactions in dynamic real-world application environments. ABMs may incur significant simulation times for large-scale business applications in the real world. A primary challenge is to reduce the simulation time while ensuring the functional correctness of the simulated application on the underlying hardware. This paper illustrates how a correct functional architecture of an application can significantly improve the application's simulation performance in terms of scalability and simulation time. We propose a metamodel to capture the communication patterns among various agents of an application deployed over ABM. This knowledge on learning communication patterns is used to design an optimally distributed software architecture for the application to minimize the number of messages communicated among agents. Further, we propose a performance model-based deployment of the architected application on the underlying hardware. We illustrate the efficacy of our approach with a use case from the telecom domain. We validate our results via detailed empirical analysis achieving a 517x gain for 1 million users in simulation time. We also scale to 50M users in barely 3.7 hours. The insights obtained from our experiments can serve as guidelines for system engineers, facilitating the process of scaling applications simulated using ABM simulators.