Kahn Process Networks and a Reactive Extension - Signal Processing Systems - page 1046

8 Reactive Process Networks

The Reactive Process Networks (RPN) model of computation intends to provide
a semantic framework and implementation model for data-flow networks with
sporadic events, in the same way KPN is a reference for determinate data-flow
models of computation. RPN integrates control and event processing with stream
processing in a unifying model of computation with a compositional operational
semantics. The model tries to find a balance in a trade-off between expressiveness,
determinism and predictability, and implementability.

8.1 Introduction

We illustrate the Reactive Process Networks model by looking at the domain of
multimedia applications, working with information streams such as audio, video or
graphics. With modern applications, these streams and their encodings can be very
dynamic. Smart compression, encoding and scalability features make these streams
less regular than they used to be.
Streams are typically parts of larger applications. Other parts of these applica-
tions tend to be control-oriented and event-driven and interact with the streaming
components. Modern (embedded) multimedia applications can often be seen as
instances of the structure depicted in Fig. 9 . At the heart of the application,
computationally intensive data operations have to be performed in streams of for
instance pixels, audio samples or video frames. Input and output of these processes
are highly regular patterns of data. These data processing activities can often be
statically analyzed and scheduled on efficient processing units. At a higher level,
modern multimedia streams show a lot of dynamism. Object-based video (de)coders
for instance work with dynamic numbers of objects that enter or leave a scene.
Decoding of the individual objects themselves uses the static data processing
functions, but they may need to be added, removed or adapted dynamically, for

Fig. 9 Embedding of types of streams