A Quality-of-Service Architecture for High-Performance Numerical Components †

We propose a model for QoS-based composition of high-performance numerical components. We define an architecture that relies on five key capabilities and services including component characterization, component proxy services, component replacement, a decision module, and archival run information processing. We describe quality metrics that are important for high-performance numerical simulations, including computational cost, accuracy, and rates of convergence and failure. We discuss the use of the architecture and quality metrics in the context of a driven cavity flow simulation, which has been shown to benefit from adaptive solution techniques that could be derived from a QoS architecture.