Static dataflow analysis for soft real-time system design

Abstract

Synchronous (deterministic) dataflow (SDF) has been extensively used to model flow constraints of digital signal processing (DSP) applications executed on (hard) real-time (RT) operating system (OS). Modern internet-of-things are, however, are often equipped with (soft) RTOSs such as embedded Linux. To reduce design iterations for the latter, the paper proposes a stochastic approach to SDF graphs where the response time of each node is modeled as probability density function (pdf). With increasing number of iterations over the graph, the individual PDFs propagate through the network. The first and second central moments of the resulting joint pdf correspond to the expected system latency and jitter, respectively. The scheduler may execute the code sequentially or in parallel. The proposed analysis tool is helpful in identifying bottlenecks within the system.