Reconfiguração dinâmica em sistemas de tempo real com restrições de consumo de energia

This dissertation presents a framework for dynamic, value-based processor time allocation in multi-modal real-time applications under schedulability and energy consumption constraints. The framework is suitable for adaptive real-time embedded systems which need guarantees of energy savings. The dynamic allocation is formulated as a discrete optimization problem for which two algorithms are proposed. The first algorithm is based on the recursive principle of dynamic programming, whereas the second is a greedy heuristic based on Lagrangian and surrogate relaxations. As a means of evaluation, the framework was applied to a real multimedia encoder. Although the formulated problem is NP-Hard, the experimental analysis has shown that the derived heuristic achieves very good approximation results with low computational cost.