Buffer allocation for real-time streaming on a multi-processor without back-pressure

The goal of buffer allocation for real-time streaming applications, modeled as dataflow graphs, is to minimize total memory consumption while reserving sufficient space for each production without overwriting any live tokens and guaranteeing the satisfaction of real-time constraints. We present a buffer allocation solution for dataflow graphs scheduled on a system without back-pressure. Our contributions are 1)We extend the available dataflow techniques by applying best-case analysis. 2) We introduce dominator based relative life-time analysis. For our benchmark set, it exhibits up to 12% savings on memory consumption compared to traditional absolute life-time analysis. 3)We investigate the effect of variation in execution times on the buffer sizes for systems without back-pressure. It turns out that reducing the variation in execution times reduces the buffer sizes. 4)We compare the buffer allocation techniques for systems with and without back-pressure. For our benchmark set, we show that the system with backpressure reduces the total memory consumption by as much as 28 % compared to the system without back-pressure. Our benchmark set includes wireless communications and multimedia applications.