Minimizing Latency and Data Memory Requirement for Real-time Chain-Structured Synchronous Dataflow

This paper considers the problem of scheduling real-time applications composed by a SDF (synchronous dataflow) chain. SDF has been wildly used in DSP (Digital Signal Processor) design environments over the past ten years. We study a SDF which is characterized by a source task, a chain of processing tasks and a sink task. The source, processing and sink tasks are executed in three different processors. Source and sink tasks are periodic real-time tasks. Processing tasks process tokens (data blocks) from the source node down to the sink node. Each task has an input and an output buffer filled with tokens. This model is typically used in embedded multimedia applications. We propose, in this paper, a scheduling algorithm which optimizes latency of the application and requires small buffer sizes. We show how to compute the latency of a SDF chain and how to fix the dimension of buffers. We then validate our results onto a multimedia system simulator of the ST Nomadik^reg platform.