Layers of Languages for Self-Modeling Systems

Abstract

Self-Modeling Systems are computing systems that have complete models of their own behavior, down to some level of detail, and that interpret those models to produce that behavior (in some applications, the interpreter itself is also modeled). Then when the system changes the models, it changes its own behavior. We have shown how our Wrappings integration infrastructure facilitates the construction, operation, and management of these systems, and the appropriate limitation of their variability. In this paper, we argue that the internal reflective processes are well-suited to representation by different languages, and that as more languages are used, each one can be simpler in definition and in its relationships to semiotically neighboring ones. Furthermore, the seeming proliferation of internal languages can be organized to have very little performance impact, since constant mappings can be made directly through partial evaluation. In computing terms, we are showing the useful separation into granularity and abstraction layers of the different kinds of activity descriptions needed in the models. We illustrate the methods and approach on CARS (Computational Architecture for Reflective Systems), a testbed for studying cooperating distributed embedded systems.