Synthesizing energy-optimal controllers for multiprocessor dataflow applications with UPPAAL STRATEGO

Abstract

Streaming applications for mobile platforms impose high demands on a system’s throughput and energy consumption. Dynamic system-level techniques have been introduced, to reduce power consumption at the expense of performance. We consider DPM (Dynamic Power Management) and DVFS (Dynamic Voltage and Frequency Scaling). The complex programming task now includes mapping and scheduling every task onto a heterogeneous multi-processor hardware platform. Moreover, DPM and DVFS parameters must be controlled, to meet all throughput constraints while minimizing the energy consumption. Previous work proposed to automate this process, by modeling streaming applications in SDF (Synchronous Data Flow), modeling the processor platform, translating both models to PTA (Priced Timed Automata, where prices model energy), and using Uppaal Cora to compute energyoptimal schedules that adhere to the throughput constraints. In this paper, we experiment with an alternative approach, based on stochastic hybrid games. We investigate the applicability of Uppaal Stratego to first synthesize a permissive controller satisfying a throughput constraint, and then select a near-optimal strategy that additionally minimizes the energy consumption. Our goal is to compare the Uppaal Cora and Uppaal Stratego approaches in terms of modeling effort, results and computation times, and to reveal potential limitations.