Implementing Adaptive Ensemble Biomolecular Applications at Scale
Many scientific problems require multiple distinct computational tasks to be executed in order to achieve a desired solution. Novel approaches focus on leveraging intermediate data to adapt the application to study larger problems, longer time scales and to engineer better fidelity in the modeling of complex phenomena. In this paper, we describe types of application adaptivity in such applications, develop abstractions to specify application adaptivity, and challenges in implementing them in software tools. We describe the design and enhancement of Ensemble Toolkit to support adaptivity, characterize the adaptivity overhead, validate the implementation of two exemplar molecular dynamics algorithms: expanded ensemble and markov state modeling, and analyze the results of running the expanded ensemble algorithm at production scale. We discuss novel computational capabilities enabled by abstractions for adaptive ensemble applications and the scientific advantages arising thereof.
READ FULL TEXT