A Case for Symbolic Limited Optimal Discrete Control: Energy Management in Reactive Data-flow Circuits
Mete Özbaltan, Nicolas Berthier
Abstract
We present a framework for achieving efficient dynamic management of configurable reactive data-flow circuits subject to global design objectives such as mutual exclusion on shared resources and minimization of energy consumption. We propose a new symbolic controller synthesis algorithm that targets the optimization of a cost function summed over a sliding window of a given number of reactions of the system. We then present a technique for constructing symbolic models of configurable data-flow circuits that lends itself to the automatic computation of dynamic configuration controllers. We use these models to experimentally evaluate our control algorithm, and make the case for symbolic optimal discrete controller synthesis on such designs.