Self-adaptive Architectures for Autonomic Computational Science

Self-adaptation enables a system to modify it’s behaviour based on changes in its operating environment. Such a system must utilize monitoring information to determine how to respond either through a systems administrator or automatically (based on policies pre-defined by an administrator) to such changes. In computational science applications that utilize distributed infrastructure (such as Computational Grids and Clouds), dealing with heterogeneity and scale of the underlying infrastructure remains a challenge. Many applications that do adapt to changes in underlying operating environments often utilize ad hoc, application-specific approaches. The aim of this work is to generalize from existing examples, and thereby lay the foundation for a framework for Autonomic Computational Science (ACS). We use two existing applications – Ensemble Kalman Filtering and Coupled Fusion Simulation – to describe a conceptual framework for ACS, consisting of

mechanisms, strategies and objectives

, and demonstrate how these concepts can be used to more effectively realize pre-defined application objectives.