Task-parallel global optimization with application to protein folding

摘要

This paper presents a software framework for high performance numerical global optimization. At the core, a runtime library implements a programming environment for irregular and adaptive task-based parallelism. Building on this, we extract and exploit the multilevel parallelism of a global optimization application that is based on numerical differentiation and Newton-based local optimizations. Our framework is used in the efficient parallelization of a real application case that concerns the protein folding problem. The experimental evaluation presents performance results of our software system on a multicore cluster.