Introduction to my Research:

Multiscale Computing

Derek Groen

Centre for Computational Science

Nature is multiscale

Multiscale modelling = combining models

of these processes, given that each has its

own scale.

Multiscale computing = doing this on a

computer, e.g. with simulation codes.

Borgdorff et al., JPDC, vol. 73,

no. 4, April 2013, pp. 465-483.

Defining multiscale computing: coupling scales

Each scale needs a model, but how can we combine them?

4 stages of

construction:

● Modelling

● Architecture

● Implementation

● Execution

Monolithic

+ Add models by extending a code.

+ Models can be integrated at a very

deep level in the code.

- The code becomes larger.

- The code becomes less portable.

- Adding new features becomes

increasingly costly.

+ Models are separate codes, and

can be swapped and updated easily.

+ Models can be run on different

computational resources.

+ Coupling models can be done

using an externally mandated ruleset.

- Coupling models may initially take a

little more effort.

Multi-model

Multiscale computing matters in a range of domains

Includes projects from

EU, NIH, NSF and

DOE.

Groen et al., IEEE CiSE, 2014.

Technologies for Code Coupling

MUSCLE 2: optimized for complex coupling

schemes, supports formalisms.Borgdorff et al., J. Comput. Sci. 5 (5), 719-731, 2014.

www.qoscosgrid.org/trac/muscle

MPWide: optimized for speed, flexibility and wide

area networks.http://www.github.com/djgroen/MPWide

Groen et al. J Open Research Software 1e9, 2013.

FabSim: optimized for using multiple remote

machines and non-concurrent coupling workflows.http://www.github.com/djgroen/FabSim

For a survey, see: Groen, Zasada & Coveney, IEEE CiSE 16 (2), 34-43, 2014.

Multiscale applications: Cerebrovascular bloodflow

HemeLB-PyNS multiscale simulations

using a desktop and a supercomputer.

3D-1D coupling.

Cyclic, concurrent.

<1% overhead for 512+4 cores.

HemeLB-Chaste multiscale

simulations using a desktop and a

supercomputer.

Fluid-Solid interactions.

Cyclic, not concurrent.

Groen et al., Interface Focus 3(2), 2013.

Bernabeu et al., Interface Focus 3(2), 2013.

Bernabeu et al., in prep.

Aim: Develop quantitative coarse-grained models of clay-polymer

nanocomposites.

Uses:

● Predict thermodynamically favourable states.

● Predict elasticity.

We require:

● Accurate potentials.

● Realistic structures.

● Task farming.

Multiscale applications: Clay-polymer nanocomposites

Suter, Groen and Coveney, Adv. Mat. 27 (6) 966-984, 2015.

Multiscale applications: Clay-polymer nanocomposites

Multiscale applications: Clay-polymer nanocomposites

● Montmorillonite clay +

Polyvinyl alcohol (PVA).

● Polymers have 30 monomers.

● Simulation box contains 8

tactoids, each with 4 platelets.

● Example: intercalation of PVA in

a clay tactoid.

Multiscale computing becomes increasingly

important to accurately model large physical

systems.

The software is there to conveniently combine

different models, and run them on large

supercomputers where needed.

Conclusions

This work was funded by the EU FP7 projects MAPPER and CRESTA, the EPSRC

funded 2020 Science project and by the Qatar National Research Foundation.

Thank you for your time

FabSim: optimized for using multiple remote machines

and running many applications at once.http://www.github.com/djgroen/FabSim

AHE:optimized for (web) applications with many users.http://www.realitygrid.org/AHE/

QosCosGrid: optimized for reserving resources and

automatically allocating jobs to them.http://www.qoscosgrid.org

Technologies for Automation

Technologies for Code Coupling

MUSCLE 2: optimized for complex coupling

schemes, supports formalisms.http://www.qoscosgrid.org/trac/muscle

MPWide: optimized for speed, flexibility and wide

area networks.http://www.github.com/djgroen/MPWide

Groen et al. JORS 1e9, 2013.

MPIMPI: widely installed, optimized for speed, but

supports only local sites.http://www.openmpi.org

For details, see: Groen, Zasada & Coveney, IEEE CiSE 16 (2), 34-43, 2014.

Multiscale Computing: from SSM to Execution

Chopard, Borgdorff and Hoekstra, Phil. Trans. R. Soc. A, 372 (2021), 2014.