Francisco Castejón [email protected] CIEMAT As coordinator of NA4-Fusion:
description
Transcript of Francisco Castejón [email protected] CIEMAT As coordinator of NA4-Fusion:
User Forum 2006
Francisco CastejónFrancisco Castejó[email protected]@ciemat.es
CIEMATCIEMAT
As coordinator of NA4-Fusion: SW-Federation (CIEMAT, BIFI, UCM, INTA -
Spain-), Russian Federation (Kurchatov Institute -
Russia-), CEA (France), ENEA (Italy), Culham (UK), Korea
(KISTI)
Fusion activities in the GRID Fusion activities in the GRID (na4-egeeII)(na4-egeeII)
User Forum 2006
OutlineOutline
Motivation: Fusion on the GRID.Data storage and handling.
Fusion Deployment and VO setup. Applications: Computing in Plasma Physics.Future Applications in the grid. Final Remarks.
Motivation: Fusion on the GRID.Data storage and handling.
Fusion Deployment and VO setup. Applications: Computing in Plasma Physics.Future Applications in the grid. Final Remarks.
User Forum 2006
MotivationMotivationLarge Nuclear Fusion installations: International
Cooperation among several Institutes. Generate ~ 1-10 GB/sec. Less than 30% of data
goes into processing. Distributed data storage and handling needed. Massive Distributed Calculation: A new way of
solving problems. (Physics Problems still without solution).
Fusion community (Science and Technology) needs new IT approaches to increase research productivity.
Large Nuclear Fusion installations: International Cooperation among several Institutes.
Generate ~ 1-10 GB/sec. Less than 30% of data goes into processing.
Distributed data storage and handling needed. Massive Distributed Calculation: A new way of
solving problems. (Physics Problems still without solution).
Fusion community (Science and Technology) needs new IT approaches to increase research productivity.
User Forum 2006
Pulse Preparation
Countdown
Pulse
Data Collection
Data Storage
Intershot Analysis
Control/Diagnostic
Setup/Validation
Offline Analysis
JET - PulseJET - PulsePulse
PreparationSimulatio
nModelling
Real-time control
Data Analysis
Data Acquisition
International ProjectRemote
ParticipationIT
SecurityNuclear Registered
Site
Data Management
Supercomputers
NA4-Catania, 2006
ITER: Making decisions in real Time !!
ITER: Making decisions in real Time !!
Data Acquisition
and Storage (Grid, Supercomputers)
Data Acquisition
and Storage (Grid, Supercomputers)
Data Analysis and Reduction: Artificial Intelligence, Neural Network, Pattern Recognition
Simulations: Large codes in different platforms (Grid, Supercomputers)
Decision for present/next shot
One half an hour shot every one and half an hour: Decisions in real time.
User Forum 2006
ITER PartnersITER Partners
Distributed Participation. Distributed Participation. Data access. Remote Control Rooms?Data access. Remote Control Rooms?
NA4-Catania, 2006
DATA HANDLINGDATA HANDLING
Storage: Large data flux: 104 sensors x 20-50 kHz sampling= 1-10 GBy per second raw datax 0.5 h= 3 TBy per shot in ITER every 1,5 h
Supercomputing and Grid Computing --> Data Storage: Scratch and permanent.
Data Access & Sharing in Large Cooperative Experiments:A unified representation for experimental data native storages that LCG-2/gLite CE middleware could understand is needed.(get and put the data files from and into the storage). Storage should allow to do some basic processing: neural network, clustering…
User Forum 2006
Communications Communications
Remote Participation tools:Data AccessLocal VisualizationVideo Conferences and ChatsRemote ControlSECURITY & ROBUSTNESS
User Forum 2006
PARTNERS and Resources for VOPARTNERS and Resources for VOSW Federation: CIEMAT, BIFI, UCM, INTA (Spain) Kurchatov (Russia). Culham Laboratory- UKAEA (UK) KISTI (South Korea). ENEA (Italy).CEA (France).
SW Federation: CIEMAT, BIFI, UCM, INTA (Spain) Kurchatov (Russia). Culham Laboratory- UKAEA (UK) KISTI (South Korea). ENEA (Italy).CEA (France).
Experience in using and developing Fusion Applications.Experience in porting applications and developing Grid Technologies.
--> Connection with EELA (Some fusion partners: Brazil, Mexico, Argentina)
User Forum 2006
VO DeploymentVO Deployment
Present: CIEMAT: 27 KSpecInts; BIFI: 8 KSpecInts; INTA: 6 nodes Present: CIEMAT: 27 KSpecInts; BIFI: 8 KSpecInts; INTA: 6 nodes
http://grid.bifi.unizar.es/egee/fusion-vo/
http://www-fusion.ciemat.es/collaboration/egee/
Within less than 6 months: JET: 38 KSpecInts; BIFI: 32 KSpecInts; CEA-Cadarache ?KISTI?, INTA?, ENEA?
Within less than 6 months: JET: 38 KSpecInts; BIFI: 32 KSpecInts; CEA-Cadarache ?KISTI?, INTA?, ENEA?
Beginning of 2007:JET: 32 adtional cores; BIFI: 32 additional cores; CIEMAT ?;CEA-Cadarache ?(second phase already committed).
Beginning of 2007:JET: 32 adtional cores; BIFI: 32 additional cores; CIEMAT ?;CEA-Cadarache ?(second phase already committed).
User Forum 2006
COMPUTING in the GRID: Present ApplicationsCOMPUTING in the GRID: Present Applications
Applications with distributed calculations: Monte Carlo, Separate estimates, …
Multiple Ray Tracing: e. g. TRUBA. Stellarator Optimization: VMEC Transport and Kinetic Theory: Monte Carlo Codes.
Applications with distributed calculations: Monte Carlo, Separate estimates, …
Multiple Ray Tracing: e. g. TRUBA. Stellarator Optimization: VMEC Transport and Kinetic Theory: Monte Carlo Codes.
User Forum 2006
Multiple Ray Tracing: TRUBAMultiple Ray Tracing: TRUBA
Single Ray (1 PE): Single Ray (1 PE): Hamiltonian Hamiltonian Ray Tracing Equations.Ray Tracing Equations.
Beam Simulation: Beam Simulation:
Bunch of rays with Bunch of rays with beam waist far beam waist far from the critical from the critical layer (100-200 layer (100-200 rays)rays)
Bunch of rays with Bunch of rays with beam waist close beam waist close to the critical layer to the critical layer (100-200 rays) x (100-200 rays) x (100-200 wave (100-200 wave numbers) numbers) ~10~1055
GRID PROBLEMGRID PROBLEM
User Forum 2006
TRUBA: Multiple Ray TracingTRUBA: Multiple Ray TracingDifferent results with the two Different results with the two approximations.approximations.
(Also useful tool for looking for (Also useful tool for looking for Optimum Launching Position Optimum Launching Position in complex devices)in complex devices)
TRUBA for EBW: TRUBA for EBW: Collaboration between IOFAN and CIEMAT. Collaboration between IOFAN and CIEMAT. Useful for all Institutes with EBW heating (Culham, Useful for all Institutes with EBW heating (Culham, Princeton, Greifswald, CIEMAT,…)Princeton, Greifswald, CIEMAT,…)
User Forum 2006
TRUBA: Multiple Ray TracingTRUBA: Multiple Ray TracingTRUBA for EBW: TRUBA for EBW: - Cylinder geometry: A single Non-relativistic ray (tens of Cylinder geometry: A single Non-relativistic ray (tens of sec.)sec.)-Real geometry in TJ-II:Coming from a supercomputer Real geometry in TJ-II:Coming from a supercomputer (VMEC).(VMEC).
- A single Non-relativistic ray (about 18’).A single Non-relativistic ray (about 18’).- A single relativistic ray (about 40’).A single relativistic ray (about 40’).
- Some problems with Geometry libraries.Some problems with Geometry libraries.
-See:See:J. L. VJ. L. Vázquez-Poleti. “Massive Ray Tracing in ázquez-Poleti. “Massive Ray Tracing in Fusion Plasmas on EGEE”. User Forum, 2006.Fusion Plasmas on EGEE”. User Forum, 2006.
User Forum 2006
TJ-II: R= 1.5 m; a= 0.2 m, M=4, l=1.
HSX: R=1.2 m; a=0.15 m
LHD: R= 3.75 m; a= 0.65 m, M=10, l=2.
CHS: R= 1 m; a= 0.2 m; M=8; l=2.
Stellarator Optimization: Choosing the best concept
User Forum 2006
Optimised Stellarators QPS and NCSXSupercomputer Optimization
NCSXQPS
User Forum 2006
The importance of Stellarator
optimization.The importance of Stellarator
optimization.
-Design of new Plasma Physics-Fusion devices under complex exigencies.
OPTIMIZATION NECESITY BASED ON KNOWLEDGE OF STELLARATOR PHYSICS.
Many Labs looking for the optimized device.
-Design of new Plasma Physics-Fusion devices under complex exigencies.
OPTIMIZATION NECESITY BASED ON KNOWLEDGE OF STELLARATOR PHYSICS.
Many Labs looking for the optimized device.
User Forum 2006
Stellarator optimization in the GridStellarator optimization in the Grid
-A lot of different Magnetic Configurations operating nowadays. OPTIMIZATION NECESITY BASED ON KNOWLEDGE OF STELLARATOR PHYSICS.Every variant computed on a separate processor (~10’) VMEC (Variational Momentum Equilibrium Code)
120 Fourier parameters are varied.
-A lot of different Magnetic Configurations operating nowadays. OPTIMIZATION NECESITY BASED ON KNOWLEDGE OF STELLARATOR PHYSICS.Every variant computed on a separate processor (~10’) VMEC (Variational Momentum Equilibrium Code)
120 Fourier parameters are varied.
rB(ψ,θ,ϕ ) =
rBm,n
m,n∑ (ψ )ei(mθ−nϕ )
R(ψ ) = Rm,nm,n∑ (ψ )cos(mθ −nϕ )
Z(ψ ) = Zm,nm,n∑ (ψ )sin(mθ −nϕ )
Coils producing field confining the plasma may be optimised numerically by variation of the field parameters.
Coils producing field confining the plasma may be optimised numerically by variation of the field parameters.
User Forum 2006
-Neoclassical Transport.- Bootstrap current.- Equilibrium vs. plasma pressure.- Stability (Balloning, Mercier,…)
-Neoclassical Transport.- Bootstrap current.- Equilibrium vs. plasma pressure.- Stability (Balloning, Mercier,…)
Partícle trajectories in W7-X
Optimization Criteria: Target Functions
-Genetic Algorithm to detect the optimum configuration for given criteria. Target Functions can be modified.
-Genetic Algorithm to detect the optimum configuration for given criteria. Target Functions can be modified.
User Forum 2006
VMEC on Kurchatov GRIDVMEC on Kurchatov GRIDLCG-2 - based Russian Data Intensive Grid consortium
resources.About 7.500 cases computed (about 1.500 was not
VMEC-computable, i.e. no equilibrium). Each case took about 20 minutes. Up to 70 simultaneous jobs running on the grid.
- See:See:V. Voznesensky. “Genetic Stellarator V. Voznesensky. “Genetic Stellarator Optimisations in Grid”. User Forum, 2006.Optimisations in Grid”. User Forum, 2006.
LCG-2 - based Russian Data Intensive Grid consortium resources.
About 7.500 cases computed (about 1.500 was not VMEC-computable, i.e. no equilibrium).
Each case took about 20 minutes. Up to 70 simultaneous jobs running on the grid.
- See:See:V. Voznesensky. “Genetic Stellarator V. Voznesensky. “Genetic Stellarator Optimisations in Grid”. User Forum, 2006.Optimisations in Grid”. User Forum, 2006.
User Forum 2006
Kinetic TransportKinetic TransportFollowing independent particle orbits
Montecarlo techniques: Particles distributed according to experimental density and ion temperature profiles (Maxwellian distribution function)
SUITABLE PROBLEM FOR CLUSTER AND GRID TECHNOLOGIES
Following independent particle orbits
Montecarlo techniques: Particles distributed according to experimental density and ion temperature profiles (Maxwellian distribution function)
SUITABLE PROBLEM FOR CLUSTER AND GRID TECHNOLOGIES
rVD =
rE ×
rB
B2 +m2q
2v2 −v⊥2( )
rB×∇
rB
B3
User Forum 2006
Kinetic TransportKinetic Transport
Example of orbit in the real 3D Example of orbit in the real 3D TJ-II Geometry (single PE).TJ-II Geometry (single PE).
~1 GBy data, 24 h x 512 PE~1 GBy data, 24 h x 512 PE
Distribution function of parallel Distribution function of parallel velocity at a given position (Data velocity at a given position (Data Analysis).Analysis).
User Forum 2006
Kinetic transportKinetic transport
No collisions: 0.5 ms of trajectory takes 1 sec. CPU..No collisions: 0.5 ms of trajectory takes 1 sec. CPU..
Collisions: 1 ms of trajectory takes 4 sec CPU. Collisions: 1 ms of trajectory takes 4 sec CPU. Particle life: 150 - 200 ms. Single particle Particle life: 150 - 200 ms. Single particle ~ 10 min.~ 10 min.
Necessary statistics for TJ-II 10Necessary statistics for TJ-II 1077 particles. particles.
NA4-Catania, 2006
Read from disk:
Input file (“input.lis”):
Number of files.Number of trajectories in each file (may be divided in blocks inside the file).
(total statistics = number of files x number of trajectories per file)Length of each trajectory.Integration algorithm and time discretization.Number of measures in time.Temperature of backgroud electrons and ions.Random seed.Initial distribution (in space and velocity space).Model used, including or not:
Collisions with backgroud ions.Electric field.T-density dependence.
Starting point according to a (space) distribution function (“perfil_densidad.in”).
Magnetic configuration (~ 20 MB file -> stored at the Fusion VO catalog) (“42_100_68.c3d”)
Magnetic field and derivatives.Magnetic coordinates -> electric field.Distance to the vaccum chamber.
Measurement times (“tiempos.lis”)
Read from disk:
Input file (“input.lis”):
Number of files.Number of trajectories in each file (may be divided in blocks inside the file).
(total statistics = number of files x number of trajectories per file)Length of each trajectory.Integration algorithm and time discretization.Number of measures in time.Temperature of backgroud electrons and ions.Random seed.Initial distribution (in space and velocity space).Model used, including or not:
Collisions with backgroud ions.Electric field.T-density dependence.
Starting point according to a (space) distribution function (“perfil_densidad.in”).
Magnetic configuration (~ 20 MB file -> stored at the Fusion VO catalog) (“42_100_68.c3d”)
Magnetic field and derivatives.Magnetic coordinates -> electric field.Distance to the vaccum chamber.
Measurement times (“tiempos.lis”)
Kinetic transport: The applicationKinetic transport: The application
User Forum 2006
Written on disk:
●“Raw” data ( “OUT????.DAT”) ●“Pictures” of the plasma at fixed times (persistence, total energy, kinetic energy, components of the velocity ...)●“Points” of escape of the particle from the TJ-II (according to several different criteria)●~ 0.9 KB / traj. (typically 1000 traj. / file -> ~ 900 KB)●writen every ~1 hour
●Histograms (“OUT????.HIS”)●Runtime histograms of serveral magnitudes (energies, componentes of velocity, fluxes ...) as a function of spacial coordinates (rho, phi, theta)●~ 0.16 KB / traj. (typically 1000 traj. / file -> ~160 KB)●writen every ~1 hour
●Debug (“RUNTIME.DAT”)●Measurements at the end of the trajectories, just for check●~ 0.2 KB / file OUT???.DAT●updated with every new OUT???.DAT (typically 1000 files -> ~200 KB)
●Individual trajectories (“ty_??_??_??.dat”) ●Complete information of each trajectory (time, position, velocity ...)●from ~10 MB to ~ GB / traj. depends on the length of the trajectory -> just for debug
Total disk space = ~ 1.3 GB
User Forum 2006
COMPUTING in the GRID: Future applicationsCOMPUTING in the GRID: Future applications
EDGE2D Application for tokamaks
Transport Analysis of multiple shots (typically 104 shots) or Predictive Transport with multiple models: e. g. ASTRA. CIEMAT(Spa) + IPP(Ger) + Kurchatov(Rus) + EFDA(UE) + …
Neutral Particle Dynamics: EIRENE:CIEMAT(Spa) + IPP(Ger)
EDGE2D Application for tokamaks
Transport Analysis of multiple shots (typically 104 shots) or Predictive Transport with multiple models: e. g. ASTRA. CIEMAT(Spa) + IPP(Ger) + Kurchatov(Rus) + EFDA(UE) + …
Neutral Particle Dynamics: EIRENE:CIEMAT(Spa) + IPP(Ger)
User Forum 2006
JET – Flagship of Worldwide Fusion: EDGE2D Equilibrium code.
JET – Flagship of Worldwide Fusion: EDGE2D Equilibrium code.
User Forum 2006
Cross section of present EU D-shaped tokamaks compared to the ITER project
EDGE2D: Determine plasma shape from Measurements: Plasma current, Pressure,
Magnetic field…
rj ×
rB=
r∇p
-EDGE2D code solves the 2 D fluid equations for the conservation of energy, momentum and particles in the plasma edge region.
-Ions, electrons and all ionisation stages of multiple species are considered.
-Interaction with the vessel walls is simulated by coupling to monte-carlo codes, to provide the neutral ion and impurity sources.
User Forum 2006
Massive Transport CalculationsMassive Transport Calculations
For Instance: Enhanced heat Confinement in TJ-II. Lower heat diffusivity for low electron density and high absorbed power density.
A different case on every PE.
0
0.5
1
1.5
2
-15 -10 -5 0 5 10 15
Pe-2562Pe-2565Pe-2558Pe-2560Pe-2559
<r> (cm)
0
0.5
1
1.5
-15 -10 -5 0 5 10 15
2562256525582560.2559
<r> (cm)0
0.2
0.4
0.6
0.8
1
1.2
-15 -10 -5 0 5 10 15
25592560255825652562
<r> (cm)
0.1
1
10
0 5 10 15
Chi-2562Chi-2565Chi-2558Chi-2560Chi-2559
r (cm)0
1
2
3
4
5
0.5 1 1.5n
0T
e-3/2 (1019 m-3 keV-3/2)
User Forum 2006
EIRENEEIRENE CodeCode
Trayectory of a He atom in TJ-II. Vertical and horizontal proyections. It starts in the green point and is absorbed in the plasma by an ionization process. The real 3D geometry of TJ-II vacuum chamber is considerd.
User Forum 2006
EIRENEEIRENE CodeCode
Two parts: 1) Following trajectories (Totally distributed) --> GRID
2) Reduction to put all together.
EIRENE Code comes from IPP (Jülich, Germany) and is extensively used by Fusion community.
0
1 109
2 109
3 109
4 109
5 109
6 109
7 109
0 0,2 0,4 0,6 0,8 1
Neutral densityHydrogen
enl13=0.65enl13=1.00enl13=1.80
enl (10
13
cm-3
)
r/a
0
5 109
1 1010
1,5 1010
2 1010
2,5 1010
3 1010
0 0,2 0,4 0,6 0,8 1
Neutral densityHelium
enl13=0.95
enl (10
13
cm-3
)
r/a
Fit
Radial profile of atoms of He in TJ-II plasmas. An average on every magnetic surface has been done
Radial profile of atoms of H in TJ-II
User Forum 2006
Final RemarksFinal Remarks
GRID technologies will enhance Fusion Research: computing and data handling.
GRID technologies will win visibility when applied to large Fusion Experiments (like ITER).
Demonstration effect: If Fusion-Grid is succesful, GRID technologies will be extensively used by Fusion Community in the future.
FIRST APPLICATIONS ARE RUNNING IN THE GRID.
GRID technologies will enhance Fusion Research: computing and data handling.
GRID technologies will win visibility when applied to large Fusion Experiments (like ITER).
Demonstration effect: If Fusion-Grid is succesful, GRID technologies will be extensively used by Fusion Community in the future.
FIRST APPLICATIONS ARE RUNNING IN THE GRID.
NA4-Catania, 2006
Para ver esta película, debedisponer de QuickTime™ y deun descompresor TIFF (sin comprimir).