A reflective middleware architecture for simulation integration

Reviewer: Scott Arthur Moody

This paper presents an alternate architectural approach for developing and integrating existing but disparate simulation environments. In contrast to existing monolithic model environments, the authors describe a new reflective middleware approach. They create new meta-models through a structural reflection process, which are then combined with a new meta-adapter framework. Their approach then supports integrating and enhancing simulations already built by domain experts, but now integrated together more seamlessly. The existing simulation environments don't have to be converted to any single framework, but can run natively-this saves on both time and maintenance. The paper includes a chart that compares this approach with other approaches, across the following dimensions: semantic objectives, domain ontologies, complexity, time management, and separation of concerns. The authors contrast their approach with existing simulation environments such as high-level architecture (HLA), which has a heavy defense department alignment in data types. They argue that even other middleware approaches, such as the common object request broker architecture (CORBA), have a similar structural dependence on existing tool data standards. In contrast, their reflective approach supports flexibility and scalability through the creation of meta-adapters-these are used to dynamically align to the types of the various execution environments. In addition, they create a series of meta-models that reflect the various simulation environments that play an integrated role. Reflection also supports time synchronization issues by aligning the various timescales of the different simulation environments. The authors describe their new approach through a case study that uses two simulators: "(a) a fire simulator that simulates the effects of fire and smoke inside a building and (b) an activity simulator that model[s] a response activity-evacuation." The authors use unified modeling language (UML) to present the meta-model. They discuss the data issues that the simulations exhibit, including how their meta-adapters will alleviate many of the traditional issues, and they show how semantic interoperability is more powerful. Aside from the US Department of Defense's complex use of HLA, this paper shows how adapting existing simulators to an HLA standard is difficult. In contrast, in a reflective architecture, each simulator can have its own data representations, internal time management, and data management. Thus, they don't require the simulators to change their implementations; the reflective approach only requires meta-model adapters. In summary, the authors describe how reflective middleware architecture can be a better integration environment when dealing with disparate simulation capabilities. They include three-dimensional (3D) reflective architecture diagrams that are very descriptive of the problem complexities. Unfortunately, the authors do not address semantic interoperability's limits and scalability challenges as the number of data formats increases with each new simulation integration. Online Computing Reviews Service