supercomputing 2013 slides
DESCRIPTION
A collection of 4 mini-presentations which I will share in the European Exascale Projects booth at SC13.TRANSCRIPT
Tools Nanomat.
Bloodflow
Multiscale Communities
Tools
Wide area message passing
Connect applications running on different platforms by establishing communication paths.Each path can be hand-tuned for better performance.More light-weight than MPI, useful for coupling and parallelizing codes over long distances.
AppsCosmological N-body simulation across supercomputers. Here MPWide facilitates wide area message passing over the wide area networks.
Apps 1D 3D
supercomputer
Multiscale bloodflow modelling. Here we use MPWide to efficiently exchange data between a desktop in London and a supercomputer in Edinburgh.
● Couples computational models.● Connects these over wide area networks
using MPWide.● Handles models’ time and space scales as
per the Multiscale Modeling and Simulation framework*.
● Supports Java, C, C++ and Fortran.● Used by 10+ production applications.
More?
http://www.github.com/djgroen/MPWidehttp://www.qoscosgrid.org/trac/muscle
Bloodflow
● Aim: Accurately model cerebrovascular bloodflow with acceptable performance.
● Approach: integrate a person-specific circulation model with a high-res local vasculature model.
● Future applications:○ Comparison of rheology models.○ Validation against medical data (ongoing).○ Look for predictive indicators of aneurysm rupture.
● And eventually predict the outcome of cerebrovascular surgery.
Example visualization
HECToR @ EPCC
SuperMUC @ LRZ
1D 3D
supercomputer
We couple the 1D Python Navier-Stokes (PyNS) solver to HemeLB to construct a multiscale model..
We use MPWide to efficiently exchange data between a desktop in London and a supercomputer in Edinburgh.
More?
Groen et al., Interface Focus 3(2), 2013.Groen et al., Journal of Computational Science 4(5), 2013.Bernabeu et al., Interface Focus 3(2), 2013.http://www.slideshare.net/DerekGroen/multiscale-modelling-of-brain-bloodflow
Nanomaterials
Aim: To develop quantitative coarse-grained models of clay-polymer nanocomposites.We will use these models to:● Predict the thermodynamically favourable state of the
composites.● Predict their elasticity.
We require:● Accurate potentials.● Realistic structures.● Task farming many MD simulations.
Atomistic representation of a charged clay sheet
Coarse-grained representation of a charged clay sheet
1 2 3
4
More?Suter, Groen, Kabalan and Coveney, MRS Proceedings 1470, 2012.Borgdorff et al., "Multiscale Simulations on distributed European e-Infrastructures", inSiDE, Vol. 10, No. 1, Spring 2012.Suter, Coveney, Anderson, Greenwell and Cliffe, Energy Environ. Sci., 2011.More papers soon. :)
Communities
Groen, Zasada, Coveney, accepted by CiSE, 2013.
Groen, Zasada, Coveney, accepted by CiSE, 2013.
Scientific Challenges
Just scratching the surface here:● Which couplings can deliver useful
information?● What information should we exchange?● How do we validate and error-check coupled
models?○ ...what if they are multi-physics as well?
Computational Challenges
● Where to simulate the models?● How do we couple?● Can we make it fast?● Can we make it reusable?● How do we analyze the resulting data?
More?
“Survey of Multiscale and Multiphysics Applications and Communities”Derek Groen, Stefan Zasada and Peter CoveneyIEEE Computing in Science & Engineering (in press), 2013.preprint at: http://arxiv.org/1208.6444
Acknowledgements
Slides made by Derek GroenThanks go out to:James Suter, Rupert Nash, James Hetherington, Peter Coveney, Hywel Carver, Stefan Zasada, Steven Rieder, Simon Portegies Zwart, Chris Kurowski, Alfons Hoekstra, Werner Dubitzky...and many others!