Optimal deployment of multiscale applications on a HPC infrastructure
PI: Bastien Chopard (University of Geneva)
Co-PIs: Costanza Bonadonna (University of Geneva), Paul Albuquerque (HES)
July 1, 2014 - December 31, 2016
Multiscale, multi-science applications are becoming more and more important in many scientific domains. This fact reflects the need to integrate many different processes into the same simulation, which often spans several order of magnitude in spatial and temporal scales. These applications are usually very demanding in terms of computational resources. High performance computers are developing fast but, still, adequate multiscale strategies must be implemented to address the numerical challenge of real life applications. Due to their multiscale and multi-science nature, these applications often require heterogeneous computing resources (e.g. a combination of a shared memory machine with a massively parallel GPU system, etc.).
In the recent FP7 European project MAPPER (www.mapper-project.eu), we have developed the so-called Multiscale Modeling and Simulation Framework (MMSF) for designing, programming, implementing and executing multiscale applications on the European computing infrastructure (EGI and PRACE). This framework has been successfully tested on applications from several fields of science and technology (fusion, computational biology, bio-engineering, nanomaterial science, hydrology) and has lead to the concept of Distributed Multiscale Applications , to indicate that both the problem and the computing infrastructure are multiscale.
The MMSF offers several benefits: a clear methodology, software and algorithm reuse, possibility to couple new and legacy codes, heterogeneous distributed computing, and access to unprecedented computing resources.
In the present project, we want to deploy the MMSF software on the Swiss computing infrastructure, and make it available to the entire scientific community. We also propose to augment the current functionality of MMSF by adding a performance evaluation method, and monitoring tools. In short, we will develop a model of computing resources that will be added to MMSF, and from which a given choice of coupled resources can be evaluated for performance, energy consumption, or utilization.
We also want to develop a new and high impact multiscale application in geoscience, using MMSF and the computing infrastructure on which it is installed. The proposed application will have a key impact on volcanic hazard assessment. It will simulate and predict the transport of volcanic ashes in the atmosphere, coupled with an aggregation solver and an explicit plume model based on fluid dynamics. Coupling all these phenomena in one single code is a challenge that has not been achieved so far. HPC resources and advance multiscale coupling technology will be essential for the success of this application.