Scenario-Aware Dataflow

Dataflow models have proven their usefulness for specifying signal processing and streaming applications. However, traditional dataflow models such as Synchronous Dataflow (SDF) and Kahn Process Networks (KPN) either lack the capability of expressing the dynamic aspects of modern streaming applications or do not support desirable analysis techniques. The dynamism often originates from various modes of operation in which resource requirements differ considerably. Such scenarios cover for example variations in the amount of data to process or variations in the functionality to perform. Neglecting dynamism may lead to unrealistic performance analysis results and therefore to dimensioning the system inefficiently. To enable obtaining more realistic performance results than with traditional design-time analysable dataflow models, a scenario-aware generalisation of SDF was recently introduced. Next to representing the data processing part of an application, this Scenario-Aware Dataflow model (SADF) also captures the control part responsible for determining the various scenarios in which the data processing part may operate. Although improving the expressive power of traditional design-time analysable dataflow models with the possibility to express (more forms of) dynamism, design-time analysis of correctness and performance remains possible. This report presents SADF in detail, including all features for specifying complex correlations between scenario occurrences. The potential of using SADF for modelling and analysing modern streaming applications is illustrated for an MPEG-4 SP decoder and an MP3 decoder.

• Scenario-Aware Dataflow
  B.D. Theelen, M.C.W. Geilen, S. Stuijk, S.V. Gheorghita, T. Basten, J.P.M. Voeten and A.H. Ghamarian.
  Technical Report ESR-2008-08, Eindhoven University of Technology, Department of Electrical Engineering, Eindhoven, The Netherlands, July 2008. (abstract, pdf)