11 Desktop Grids for International Scientific Collaboration International Desktop Grid Federation APPLICATION OF DESKTOP GRID TECHNOLOGY IN MATERIAL SCIENCE A.Gatsenko, A.Baskova, Yu.Gordienko G.V.Kurdyumov Institute for Metal Physics, National Academy of Sciences of Ukraine, Kiev, Ukraine
22 Simulation of structure and mechanical properties of materials is extremely important in materials science to quantify deformation and strength characteristics of materials. Among variety of new materials the special place is occupied by materials that have nanoscale structure (nanomaterials), such as metal nanocrystals and nanoscale non- metallic materials with unique properties (nanotubes, graphene).
33 Problem Molecular dynamics (MD) simulations for realistic configurations take: huge resources of supercomputers large shared memory big number of CPUs.
44 Way to Solution The distributed computing (DC) model on the basis of: -BOINC, -XtremWeb-HEP, -OurGrid, -EDGeS-bridge, -WS-PGRADE
55 Main Aim To demonstrate the capabilities of the proposed technical solutions to the example of modeling of physical processes: tension of nanocrystals in different conditions tension ensemble of nanocrystals simulation graphene
66 Technical solution : Open- Source Simulator- LAMMPS Technical solution : Open- Source Simulator - LAMMPS LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator – by Sandia National Laboratories with: scripts for pre- and post-processing, multi-core CPU and GPU support, checkpointing support; intrinsic message passing, NO explicit DCI support! Very popular: numerous users/publications --> LAMMPS
77 Porting to DCI on Desktop Grid LAMMPS
88 Operational production infrastructure IMP team maintained and scaled-up DG BOINC infrastructure at the premises of IMP Desktop Grid ( From December 10, 2010 it works on the permanent basis with public access of workers (users).
99 Operational production infrastructure The current status of IMP Desktop Grid infrastructure is “under tests of scaling-up” and “under tests of new applications”. DG ( was scaled-up from 1500 to > 3000 workers (users); from to in- progress workunits. The current average performance is ~150 GFLOPs with weekly peaks of 550 GFLOPs Last 4-weak performanceNumber of in-progress workunits
10 10 Technical Solution - Conclusions Using the Desktop Grid technology with the assistance of volunteer computing resources quickly and easily achieve the required level of performance.
11 11 Demonstrations for different physical processes: single nanocrystals under different tension conditions, ensemble of nanocrystals under tension, simulation of graphene under different tension conditions
12 12 Physical process 1: Tension of nanocrystals in different parameters (conditions)
13 13 Simulation of nano-crystal Al: Typical Sweeping Parameters… External mechanical influence with different values of increasing strain… strain
14 14 Simulation of nano-crystal Al: Typical Sweeping Parameters… External mechanical influence with different crystal orientations… strain
15 15 Video of Al nano crystal strain
16 16 Simulation of nano-crystal Al: Typical Sweeping Parameters… External mechanical influence with different values of rate…
17 17 Practical Results Physical parameter decomposition for “parameter sweeping” parallelism allow us to widen a range of simulated parameters and find their “magic” (critical) values for atomic self-organization… 3D hierarchic network of voids in Al bulk 2D super-lattice on Al surface
18 18 Physical Process 1 - Conclusions the estimation of the influence of various parameters on the process of deformation of materials. the regimes allow you to create a given structure.
19 19 Physical process 2: Tension ensemble of nanocrystals with different statistical realizations
20 20 …….. Fitting PDF and CDF to Weibull distribution Distribution (PDF) of concentrations of defects in the ensemble of ~1000 samples Drift of PDF (from normal to Weibull) in ensemble of ~1000 samples: quantity->qualitative change Simulation of nano-crystal Al: Statistical analysis
21 21 Physical Process 2 - Conclusions Change of defect distribution (from normal to Weibull) is followed by qualitative change of plastic deformation mode (from homogeneous strain to localized mode and… fracture!).
Physical process 3: Simulation of graphene for different parameters
Simulation of graphene 23 23
24 24 Simulation of graphene plates Typical Sweeping Parameters… The size effect for different sizes of plates… from 2x2 nm to 2x32 nm
25 25 Simulation of graphene plates Typical Sweeping Parameters… “Tersoff” “Airebo” influence of the type of interatomic potential… strain
26 26 stress dependence of the tensile strain fracture Practical Results
27 27 Physical Process 3 - Conclusions Qualitative and quantitative analysis of the process of deformation and fracture of graphene occurs fragile scenario without the formation of stable defect substructure. A comparative analysis of the effect of different potentials (Airebo / Tersoff) on qualitative and quantitative process of deformation of graphene.
28 28 General Conclusions It is shown that the mechanical characteristics evaluated on the basis of MD simulations using LAMMPS package in the DG-SG DCI are in satisfactory agreement with the experimental data and allowed to discover the new aspects of deformation and fracture mechanisms in nanomaterials porting MD-applications to DG-SG DCI is easy, if: BOINC SZTAKI DC-API and SG-DG EDGeS Bridge are used; parameter decomposition and sweeping parallelism is possible; message passing is localized at worker side (in multi core CPU/GPU).
29 29 Acknowledgements This work is partially funded by FP7 DEGISCO (Desktop Grids for International Scientific Collaboration) ( DEGISCO project is supported by the FP7 Capacities Programme under grant agreement number RI Thank you for your attention!