Computational Seismology and Grid Computing – Application...
Transcript of Computational Seismology and Grid Computing – Application...
Computational Seismology and Grid
Computing – Application and Potential
Computational Seismology and Grid Computational Seismology and Grid
Computing Computing –– Application and PotentialApplication and Potential
ShiannShiann--Jong LeeJong Lee
Institute of Earth Sciences Academia Sinica
ISGC2008 2008/04/10
OutlineOutline
High performance computing in Earth SciencesHigh performance computing in Earth SciencesComputing, Visualization and StorageComputing, Visualization and Storage
Computational seismologyComputational seismology
Examples from earthquake source, path and site studiesExamples from earthquake source, path and site studies
Application and potential of Gird computing in Application and potential of Gird computing in seismologyseismology
High performance computing in Computational Seismology
High performance computing in High performance computing in Computational SeismologyComputational Seismology
Why high performance computingWhy high performance computing
-- ComputationComputation
-- Visualization Visualization
-- StorageStorage
HPC in Computational HPC in Computational SeismologySeismology
-- Earth Simulator (Japan)Earth Simulator (Japan)
-- Caltech GPS Dell Cluster (USA)Caltech GPS Dell Cluster (USA)
-- ERI SGI ERI SGI AltixAltix System (Japan)System (Japan)
Computation
Visualization
Storage
The Earth Simulator (2002)The Earth Simulator (2002)The Earth Simulator (2002)http://www.es.jamstec.go.jp/index.en.html
Caltech GPS Dell Cluster (2006)Caltech GPS Dell Cluster (2006)Caltech GPS Dell Cluster (2006)http://citerra.caltech.edu/wiki/
- 512 dual-processor quad-core nodes
- 4096 MPI processes
- 6144 Gb memory
ERI SGI Altix System (2003)ERI SGI ERI SGI AltixAltix System (2003)System (2003)http://wwweic.eri.u-tokyo.ac.jp/computer/
Computational SeismologyComputational SeismologyComputational Seismology
Source StudiesSource Studies-- RealReal--time Gridtime Grid--based CMT: based CMT: distributed computingdistributed computing-- FiniteFinite--fault source inversion: fault source inversion: parallel computing, parallel computing, storage,storage, visualizationvisualization
Path StudiesPath Studies-- FiniteFinite--frequency tomography study: frequency tomography study: parallel computing, parallel computing, storagestorage-- GreenGreen’’s function database: s function database: storagestorage
Site and Comprehensive SimulationSite and Comprehensive Simulation-- 33--D, full waveform modeling: D, full waveform modeling: parallel computingparallel computing, , visualizationvisualization-- RealReal--time analysis: time analysis: high performance computing, high performance computing, storagestorage-- Hazard analysis, earthquake database: Hazard analysis, earthquake database: high performance computing, high performance computing, storagestorage
• Parallel computing is the simultaneous use of multiple compute resources to solve a computational problem. • Distributed computing is a science which solves a large problem by giving small parts of the problem to many computers to solve and then combining the solutions for the parts into a solution for the problem.
Strong Ground Motion Simulation of the Strong Ground Motion Simulation of the 1999 Chi1999 Chi--Chi, Taiwan, EarthquakeChi, Taiwan, Earthquake
Inversion resultsInversion results
0
0.5
1
2
3
12
6
Slip(m
)
Slip distribution Rupture process
WaveWave--Field SnapshotField Snapshot
Synthetic vs. ObservationSynthetic vs. Observation
Numerical modeling of seismic wave Numerical modeling of seismic wave propagation in the Taipei basinpropagation in the Taipei basin
1999, Chi-Chi earthquake (Mw 7.6)
2002, 331 earthquake (Mw 7.1)
1986, Hualien offshore earthquake (Mw 7.3)
(a) Map view of the Taipei basin. The depth of the basement is represented by gray colors. The red line shows the JhongShan freeway across the basin. The location of the world’s current tallest building, Taipei 101, is indicated in the eastern part of the basin. (b) Perspective view of the two major discontinuities in the Taipei basin: The SongShan formation and the basin basement. Surface topography around the basin is shown at the top of the figure.
Taipei BasinTaipei Basin
North Taiwan SEM MeshNorth Taiwan SEM Mesh
Realistic Topography
Taipei basin mesh
Caltech's Division of Geological & Planetary Sciences Dell cluster
512 dual-processor quad-core nodes
Caltech's Division of Geological & Planetary Sciences Dell cluster
512 dual-processor quad-core nodes
TopographyBasinMoho3D Velocity
2004/10/23 Taipei Earthquake (M2004/10/23 Taipei Earthquake (MLL 3.8)3.8)
PGA SimulationSynthetic vs. Observation
Velocity waveformBand-pass filtered between 0.8 and 10 sec
Computational VisualizationComputational VisualizationSouthern California Earthquake Center (SCEC) – SDSC Visualization Services
The Earth Simulator Center - Atmosphere & Ocean Simulation Research Group
We carry out simulation researches using CFES (CGCM for the Earth Simulator) to understand the mechanism of the variability and to study the predictability in the coupled atmosphere–ocean system.
The TeraShake simulations modeled the earth shaking that would rattle Southern California if a 230 kilometer section of the San Andreas fault ruptured producing a magnitude 7.7 earthquake.
TopographyBasinMoho3D Velocity
100km
102km
88km
Taipei BasinN
Doublet event
I-Lan Doublet Event2005/03/06, ML = 5.9
! resolution of the mesh at the surface:! -------------------------------------!! spectral elements along X = 448! spectral elements along Y = 864! GLL points along X = 1793! GLL points along Y = 3457! average distance between points along X in m = 116.8700 ! average distance between points along Y in m = 109.9049 fast
slow
Seism
ic Wave V
elocity
NN380 km
210 km
100 km
? ?
??
Coastal Range
Longitudinal Valley
Central RangeWestern Plain
PHILIPPINE
SEA PLATE
EURASIAN
PLATE
Community mesh model Community mesh model for the whole Taiwanfor the whole Taiwan
Moho
HPC Cluster in IESHPC Cluster in IESHPC Cluster in IESIBM Blade Server : 20 MPI processes (2004) IBM Blade Server : 20 MPI processes (2004) PC Cluster: 32 MPI processes (2007)PC Cluster: 32 MPI processes (2007)
WhatWhat’’s Grid Computing?s Grid Computing?
Application and potential of Application and potential of Gird computing in seismologyGird computing in seismology
WhatWhat’’s Grid Computing?s Grid Computing?
Result outputResult output
• Rupture model (source)
• Simulation region (Path and Site)
• Physical properties (Maximum frequency, Minimum Velocity and so on)
Problem definition
InputInput
Community models
Grid computing
ASGC Grid Resource
Data Grid
ComputingComputing
StorageStorage
IES or elsewhere
GridGrid--based Computing Pathwaybased Computing Pathway
Hazard map
Numerical visualization
Visualization, AnalysisMachines (ASGC, IES)
NumericalNumericaloutputoutput
GridGrid--based Visualization Frameworkbased Visualization Framework
Reduction
TCP/IP
Rendering
Sorting Transport
... ... ... ...
Receive Buffer
Parallel I/O
Grid Resources Visualization Machine(modified from SCEC CME project)
4D visualization of Oct. 23, 2004Taipei earthquake
SummarySummaryHigh performance computing have succeeded in applying to High performance computing have succeeded in applying to seismology, such as source, path, site effect studies and seismology, such as source, path, site effect studies and comprehensive 3comprehensive 3--D simulation.D simulation.
However, constructing a realistic earthquake simulation from However, constructing a realistic earthquake simulation from source and path models of constituent phenomena and source and path models of constituent phenomena and executing that simulation on suitable computing platforms executing that simulation on suitable computing platforms becomes increasingly complex.becomes increasingly complex.
We are finding possible collaborations between researchers in We are finding possible collaborations between researchers in the information technology areas and earthquake scientists to the information technology areas and earthquake scientists to deal with more and more complex seismologic problems. deal with more and more complex seismologic problems.
The Grid technique is therefore one of the best candidate for The Grid technique is therefore one of the best candidate for computational seismology studies in the near future.computational seismology studies in the near future.
4D Visualization of the 1999 Chi-Chi, Taiwan, Earthquake Mw 7.6
For more information: http://www.earth.sinica.edu.tw/~sjlee/index.htm
Thank you