Abstract Domain Interaction and Communication in Ptolemy II

Researchers: Stephen Neuendorffer
Advisor:Edward A. Lee

Many models of computation have been designed to allow the description of different kinds of systems. For example, Synchronous/Reactive and Finite State Machine models have been shown to be appropriate for control-oriented problems, while Dataflow models work well for modeling signal processing systems. Furthermore, certain dataflow models of computation can be shown to have nice properties, such as Turing decidability, determinacy, and schedulability. However, few systems are easily described using only one model of computation. The principle of "the right tool for the right job" suggests that different parts of the system should be described using the most appropriate model of computation. Such systems are called heterogenous systems. This research is focused on extracting the basic semantics of different models of computation in order to provide mechanisms within Ptolemy II for simulating heterogenous systems.

Much research has been involved with extending one model of computation to handle a broad range of problems. However, this approach tends to run into problems, since broadening the scope of a model of computation generally sacrifices the useful properties that can be proved about a more limited model. Ptolemy II aims to provide an architecture which is capable of modeling a wide range of problems, by allowing different domains (which encapsulate a model of computation within an abstract interface) to interact. This interaction is restricted to a clustered graph syntax, where the entities within a domain interact according to the semantics of that domain, but each entity may contain a domain of its own that operates under completely independant semantics.

In Ptolemy 0.7.1, many assumptions were made in the kernel that were specific to dataflow oriented domains. This fact made the implementation of some domains, such as Finite State Machines, very difficult. In Ptolemy II, a great effort has been made to abstract the kernel so that no assumptions are made that may dictate a particular model of computation. Several domains have already been implemented within this framework, which appears to work well as an abstract basis for a wide range of models of computation. This research is intended to explore the interactions between domains and to discover an abstract model of communication between domains that preserves the properties of each domain. If successful, this will allow Ptolemy II to specify intuitive semantics of systems described using multiple domains, and allow more complete representations of problems.

Last updated 11/22/98. Send comments to www@ptolemy.eecs.berkeley.edu.