Parameterized Dataflow Scenarios

Although well-suited for capturing concurrency in streaming applications, purely dataflow-based models of computation are lacking in expressing intricate control requirements that many modern streaming applications have. Consequently, a number of modeling approaches com-bining dataflow and finite-state machines has been proposed. However, these FSM/dataflow hybrids struggle with capturing the fine-grained data-dependent dynamics of modern streaming applications. In this article, we enrich the set of such FSM/dataflow hybrids with a novel formalism that uses parameterized dataflow as the concurrency model. We call the model FSM-based parameterized scenario-aware dataflow (PFSM-SADF). Through the use of parameterized da-taflow, the formalism can capture the application fine-grained data-dependent dynamics while the enveloping FSM enables the capturing of the application control flow. We demonstrate the application of our modeling framework to synchronous dataflow (SDF), for which we propose a worst-case performance analysis framework based on the Max-plus algebraic semantics of SDF and the theory of Max-plus automata. We show that using the novel hybrid one can give tighter bounds on worst-case performance metrics such as throughput and latency for streaming appli-cations exposing fine-grained dynamic behavior embedded inside a control-flow structure then by using the existing hybrids. We evaluate our approach on a realistic case-study from the multimedia domain