An accurate energy model for streaming applications mapped on MPSoC platforms

Abstract

In this paper, we propose a very accurate energy model for streaming applications modeled as Polyhedral Process Networks (PPN) and mapped onto tile-based MPSoC platforms with distributed memory. The energy model is based on the well-defined properties of the PPN application model. To guarantee the accuracy of the energy model, values of important model parameters are obtained by real measurements. The proposed energy model is applicable to different types of processors and communication infrastructures within an MPSoC platform. The energy model is evaluated on FPGA-based MPSoC platforms against real measurements of energy consumption from the FPGA. The obtained energy estimates are highly accurate with an average error of 4% and a standard deviation of 3%. The average model evaluation time per design point takes 2.5 minutes for considered cases, which is very good given the high accuracy of the model.