Modeling control tokens for composition of CAL actors

Dataflow programming is typically used as an intuitive representation for describing multimedia and signal processing applications as computation nodes which communicate through FIFO queues. To run a dataflow network, consisting of several nodes, either run-time or compile-time scheduling is required. Compile-time scheduling techniques are typically based on token rates between nodes and for languages such as CAL, which are expressive enough to describe an actor with any behavior, run-time scheduling is needed in the general case, introducing an overhead. However, the well defined structure of dataflow programs enables analysis of the dependencies of the program and partitions with piecewise static schedules can be derived. In this paper we describe how actor partitions with control tokens can be modeled such that a correct scheduler, where most scheduling decisions are taken at compile-time, can be derived for the resulting composed actor.