Maximally Permissive Composition of Actors in Ptolemy II.

摘要

The 'Cyber' and 'Physical' worlds are merging. Cyber-Physical Systems (CPS) are no longer isolated, but start to reach into the Cloud, thereby composing a network which realizes the concept that became known as The Internet of Things. The dynamic nature of the applications in this domain poses significant technical challenges concerning the assurance of important system properties like reliability, robustness, adaptability, and security. Modeling has proven itself to be a valuable tool in gaining better understanding of complex systems, but existing modeling platforms may lack the expressivity to model these new, much more dynamic, and opportunistically composed systems in which the data they handle typically does not conform to a rigid structure. This thesis addresses the problem of handling dynamic data, in the statically typed, actor-oriented modeling environment called Ptolemy II. It explores the possibilities of using type inference to statically type dynamic data and leverage dynamic type checking to invoke error handling strategies that enhance robustness. The goal is to achieve maximally permissive composition, and the presented solution comes in the form of backward type inference. Backward inferred types are specific enough not to limit composability and general enough not to impose unnecessary constraints on the data. The type constraints imposed by downstream actors determine the type of the otherwise underdetermined output ports of actors that mediate access to untyped resources. This is achieved using additional type constraints, without changing Ptolemy II's original type resolution algorithm, and with no significant impact on the run-time of type resolution. The proposed solution was implemented successfully and has been adopted as an extension of the Ptolemy II type system. As a byproduct, this thesis gives a thorough case study of the Ptolemy II type system.