Simulation Design: Trans-Paradigm Best-Practice from Software Engineering

There are growing initiatives to apply software engineering (SE) best-practice to computational science, which includes simulation. One area where the simulation literature appears to be particularly light is in the overall structural design of simulations, and what architectures and features are valuable for what reasons. (Part of the problem is that parts of this knowledge are abstracted away in simulation toolkits which are often not easily comparable, and have different conceptual aims.) To address this, I outline three key software properties which embody SE best-practices, and then define an 'idealised' software architecture for simulation—what SE would call a reference architecture—which strongly exhibits them. I show that this is universal to all simulations (largely because modelling-paradigm-specific detail is encapsulated into a 'single black box' layer of functionality) but that simulation toolkits tend to differ in how they map to them; this relates to the aims of the toolkits, which I provide a useful categorisation of. I show that, interestingly, there are several core features of this architecture that are not fully represented in any simulation toolkit that I am aware of. I present a library—JSIT—which provides some proof-of-concept implementations of them for Java-based toolkits. This library, and other ideas in the reference architecture, are put into practice on a published, multi-paradigm model of health and social care which uses the AnyLogic toolkit. I conclude with some thoughts on why this area receives so little focus, how to take it forwards, and some of the related cultural issues.