real world applications on massively parallel environments · •optimization in macroscopic...
TRANSCRIPT
Real World Applications on Massively Parallel Environments
YAMAZAKI, Takashi, Ph.D.
Application R&D Division
Next Generation Technical Computing Unit
Fujitsu Ltd. Copyright 2012 FUJITSU LIMITED
SC12 Nov. 10-16th, 2012
Contents of Presentation
Copyright 2012 FUJITSU LIMITED
Introduction
Industry: Electromagnetic field simulator
Environment: Tsunami simulator
Healthcare : Heart simulator
Conclusion
1
Introduction
Copyright 2012 FUJITSU LIMITED
Application Areas
Health care - aid diagnosis - personalized therapy - efficient drug trial
Industry - new product - new material - energy saving
Environment - prediction and prevention of disaster - prepare for and protect against disaster
Aims
Effective utilization of HPC
Broaden the possibilities of HPC
Practical applications
2
Copyright 2012 FUJITSU LIMITED
i. Design of Motor
ii. Efficiency Improvement of Magnet
Industry: Electromagnetic field simulator
3
Energy Saving Design of Motor
Motors consume almost half of the generated electric power
The same tendency is also observed globally
Copyright 2012 FUJITSU LIMITED
motor 56%
light 14%
heater 10%
IT 5%
other 15%
in Japan (2009, Fuji-Keizai)
570 TWh
Saving One thermal plant
1% more efficient
4
Coupling of Simulation of Two Scales
Copyright 2012 FUJITSU LIMITED
magnetization
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
-100 -50 0 50 100
Magnetic Field (A/m)
Mx/M
s
hysteresis loop
weakly coupling
micro-magnetics hysteresis model (material property)
EM analysis: FEM (deformation & current)
Highly precise magnetic field analysis for design of products realized
FEM analysis result
5
Rare Earth Magnet Analysis
Magnetic reversal* process of rare earth magnet analyzed with 3-D micromagnetics simulation
8~10 nm
Copyright 2012 FUJITSU LIMITED
mesh size < 1nm
*magnetic reversal: the phenomenon which direction of magnetization reverses by direc-tion of an external magnetic field.
6
Copyright 2012 FUJITSU LIMITED
This whole project is a collaboration with Prof. Imamura of IRIDeS, Tohoku University
Environment: Tsunami simulator
7
Prevention of Tsunami Disaster
Copyright 2012 FUJITSU LIMITED
The Digital terrain data is provided by
GEO Technology Laboratory Co., Ltd.
Mar., 11th, 2011
M9 earthquake & historic tsunami
Over 19,000 victims
Target
Tsunami-proof design of buildings
Escape planning
Preliminarily enlightening on the danger
City planning
Tolerability of evacuation cites or public buildings
This movie is for the demonstration of the simulation technique,
not for prediction of actual tsunami damage.
8
For Safety Urban Planning
Copyright 2012 FUJITSU LIMITED
1. Calculation of wave stress
2. Inundation analysis w/o overdamping
Formulation of tsunami-proof intensity standard
Tsunami-proof design
Re-examination of dikes
Escape planning
Hazard analysis
Preliminarily enlightening on the danger
Tolerability of evacuation cite or public building
9
Breaking Wave on Surface
Copyright 2012 FUJITSU LIMITED
caseA4 W6
-1
-0.5
0
0.5
1
1.5
2
30 31 32 33 34 35 36
SPH 水槽実験
The time series of wave height matches the experimental data.
The source of the experimental data is Oki, Murakami, Mase (2002) Proc. Coast. Eng., 49, 41.
Surface motion at an artificial reef
wave breaking wave reforming
simulation
The hydraulic experime-nt is performed by Prof. Murakami of Univ. of Miyazaki.
10
Combination of 2-D & 3-D Simulations
Seamless 3-D simulation from the epicenter to an urban area
Copyright 2012 FUJITSU LIMITED
tsunami propagation. from epicenter inundation in urban areas
Shallow-water model (2-D) for broad area
SPH method (3-D) for small area
combine
tsunami simulator
Simulation performed by Porf. Imamura of Tohoku Univ. CG made by the collaboration with Bosai Tech Consultants.
The Digital terrain data is provided by GEO Technology Laboratory Co., Ltd.
10m 1km 100km
calculation of wave stress to buildings
11
Copyright 2012 FUJITSU LIMITED
This work is a collaboration with Prof. Hisada and Prof. Sugiura of the University of Tokyo and Fujitsu Limited
Healthcare: Heart simulator [UT-Heart]
12
For Personalized Therapy
CVD (cardiovascular disease)
One of the leading causes of death
Computer aided therapy
Personalized therapy
Efficient drug discovery
Copyright 2012 FUJITSU LIMITED 13
Multiscale & Multiphysics
Copyright 2012 FUJITSU LIMITED
Mu
ltip
hys
ics
mechanics
Bioelectricity
Biochemistry
cross bridge
Action
potential ECG
P-V loop
Tension
Glucose
glycolysis ATP
Lactate
sarcomere
Multiscale
molecule tissue organ human Cell
10-9m 10-4m 10-2m 100m
cardiac myocyte
K+ Na+ Ca++
Membrane
protein
14
Copyright 2012 FUJITSU LIMITED
Electrocardiogram (ECG)
Lead II ECG Cardiac Excitation
Comparison of ECG measurement between heart simulator and human heart
simulated
Potential Distribution on torso surface
observed
15
Ventriculoplasty
Virtual surgical treatment
Copyright 2012 FUJITSU LIMITED
EF: Ejection Fraction. EF represents the strength of pumping function of heart. Defined as (end-diastolic volume – end-systolic volume) / (end-diastolic volume).
Diseased heart Virtual operation Following virtual therapy
40 60 80 100 120 1400
5
10
:large cut:middle cut:myocardial infarction
Left Ventricular Volume[cc]
Lef
t V
entr
icul
ar P
ress
ure
[KP
a]
---
EF
①
②
③
34.9
46.6
50.9
Ventricular aneurysm after myocardial infarction
pressure-volume looop
modest
radical
excision sutura
②
③
①
16
EC-coupling metabolism, sarcomere dynamics
structural dynamics by FEM contraction force
FSI by FEM circulation
Multiscale Analysis
molecule cell tissue organ
homogenization method
micro macro
For the personalized therapy in medical domain
As a bridge between molecular biology and clinical application
multiscale simulation
Copyright 2012 FUJITSU LIMITED 17
Scalability and Performance on "K computer"
Succeeded in simulating 1.5 heartbeats lasting about 17 hours and using 602,112 cores.
Achieved more than 90% of the ideal scaling and 27.7% of the peak performance using 20,736/41,472/82,944 nodes with 659,456 embedded numerical cells of and 49,248 DOFs.
Copyright 2012 FUJITSU LIMITED
0
1
2
3
4
0 50000 100000
Spee
d-u
p r
ati
o
#Nodes
Ideal
49k DOF
#Nodes 20,736 41,472 82,944
Pflops 0.78 1.49 2.94
% of the peak performance
29.32 28.12 27.72
% of the ideal scaling
100 98.0 95.3
close to the ideal slope
Present results were obtained by early access to the "K computer" at RIKEN AICS as a "grand challenge application”
under the "R&D of Next-Generation Integrated Life-Science Simulation Software" program supported by MEXT.
strong scalability
18
Conclusion FUJITSU has been developing the HPC applications for practical use
Effective utilization of HPC
Broaden the possibilities of HPC
Practical applications
Some of these were introduced.
Tsunami simulator based on article method for disaster prevention
• From the epicenter to the urban area
FEM based Electromagnetic field simulator for industry
• Coupled with micromagnetics simulator
• Optimization in macroscopic quantities possible, taking into account of the effect of micro magnetics
FEM based human heart simulator (UT-Heart) for personalized medical therapy
Multiscale & multiphysics heart simulator
Possibilities of use to aid diagnosis not only in surgery but also in a medical domain for personalized medical treatment
Achieved more than 90% of the ideal scaling and 27.7% of the peak performance on "K computer"
Copyright 2012 FUJITSU LIMITED 19
Copyright 2010 FUJITSU LIMITED
Thank you for your attention. 20