Design for Predictability and Efficiency (PREDATOR) Reinhard Wilhelm Saarland University Saarbrücken, Germany
Outline Objectives of PREDATOR Overall Structure of PREDATOR The Work Packages Starting Points of our Work Interactions between WPs Milestones and deliverables Expected activities of PREDATOR Project management
Motivation Design of Real-Time Systems: Gap average-case worst-case behavior increases Static analyses: derive guarantees about worst-case behavior Tightness of guarantees: Product of Uncertainty x Penalties Predictability vs. Efficiency: Past systems were… either predictable, but inefficient (no caches, pipelines, over-provisioning of resources, …), or efficient, but unpredictable (caches, deep pipelines, branch prediction, speculation, …) PREDATOR: Reconcile Predictability and Efficiency
Objectives of PREDATOR Overall Goals of PREDATOR: Reduce Uncertainty by increasing system analyzability Reduce Penalties by influencing system design Objectives: Improve design & development methods for real-time systems Develop tools supporting such development methods Develop architectural concepts supporting timing analysis Approach: Synergetic development of tools with design Predictability vs. efficiency: Multi-objective optimization problem Resource-aware abstraction during system design
PREDATOR Partners Saarland University (Coord.): R. Wilhelm ETH Zürich: L. Thiele TU Dortmund: P. Marwedel Alma Mater Stud.-Università di Bologna: L. Benini Scuola Superiore Sant’ Anna, Pisa: G. Buttazzo AbsInt GmbH, Saarbrücken: C. Ferdinand EADS Airbus, Toulouse: B. Triquet Robert Bosch GmbH, Stuttgart: O. Rogalla
Workpackages WP5 – Cross-Layer Design & Analysis WP4 – Distribution and MPSoC WP0 – Use Cases WP3 – Coordination between Tasks WP6 – Dissemination WP7 – Management WP2 – Single-Task Layer WP1 – Hardware Architecture