Fusion Simulations, data visualization Fusion Simulations, data visualization and future requirements for the and future requirements for the interactive grid infraestructureinteractive grid infraestructure

F. CastejónF. Castejón11, D. López Bruna, D. López Bruna11, J.M. Reynolds, J.M. Reynolds22, A. Tarancón, A. Tarancón22, , R. VallésR. Vallés22, J.L. Velasco, J.L. Velasco22

1.1. Instituto de Biocomputación y Física de Sistemas ComplejosInstituto de Biocomputación y Física de Sistemas Complejos2.2. Laboratorio nacional de Fusión-Asociación Euratom/Ciemat Laboratorio nacional de Fusión-Asociación Euratom/Ciemat

28040 Madrid, Spain.28040 Madrid, Spain.

WWW on FusionWWW on Fusion What is Fusion?What is Fusion?

• Increasing the temperature of a gas, we get a plasma stateIncreasing the temperature of a gas, we get a plasma state• At this temperature, the union of light atom nuclei is possible At this temperature, the union of light atom nuclei is possible

through an exothermal process:through an exothermal process: Mass after fusion process less than before itMass after fusion process less than before it Exceeding mass -> energyExceeding mass -> energy

Why Fusion?Why Fusion?• Long-term safe and enviromentally friendlyLong-term safe and enviromentally friendly• Delocalized fuel: water and lithiumDelocalized fuel: water and lithium

When Fusion?When Fusion?• ITER will produce about 400MW of power excess, and it is the ITER will produce about 400MW of power excess, and it is the

previous step to DEMO, a demostration reactor, the last stage previous step to DEMO, a demostration reactor, the last stage (50 years)(50 years)

High

TemperaturesGAS PLASMA

THEORETICAL MODELSTHEORETICAL MODELS

PLASMA SIMULATIONPLASMA SIMULATION Why?Why?

• Reach a better understanding of plasma physics is Reach a better understanding of plasma physics is necessary to obtain good modelsnecessary to obtain good models

• Not expensive, compared with the experimentNot expensive, compared with the experiment• Very FlexibleVery Flexible

What?What?• Under several non-restrictive assumptions, the Under several non-restrictive assumptions, the

plasma is modeled by the equations:plasma is modeled by the equations:

• Guiding center aproximationGuiding center aproximation• Evolution of the distribution function for every specie:Evolution of the distribution function for every specie:

Convection termsConvection terms Collision termsCollision terms

Solving GC plasma equationSolving GC plasma equation The parameters are:The parameters are:

• Magnetic field (supose fixed)Magnetic field (supose fixed)• Electric field (for the moment supose fixed)Electric field (for the moment supose fixed)

The only aditional non linearity are the collision The only aditional non linearity are the collision termsterms

Even assuming all previous simplifications, it is Even assuming all previous simplifications, it is not possible to obtain a general analytical not possible to obtain a general analytical solution in the 5D space.solution in the 5D space.

It is neccesary to obtain numerical solutions. Two It is neccesary to obtain numerical solutions. Two possible ways are:possible ways are:• Langevin approachLangevin approach• Direct approachDirect approach

Linealizing the collision term:Linealizing the collision term:• Equivalent to applying a thermal Equivalent to applying a thermal

bathbath• Then, equation has a Focker-Then, equation has a Focker-

Planck equivalent:Planck equivalent:

• This is, the evolution of the distribution function is equivalent to solve many independent test particles trajectories collisioning with a fixed background caracterized through noise terms.

Relatively easy to gridify (no communication between particles)

Langevin Approach: ISDEPLangevin Approach: ISDEP

Solve directly the equation to get the distribution Solve directly the equation to get the distribution function:function:• Get a magnetic geometry adapted meshing in 3D spatial Get a magnetic geometry adapted meshing in 3D spatial

dimensions.dimensions.• Get a truncated spectral expansion of the DF in space and Get a truncated spectral expansion of the DF in space and

velocity:velocity:

If mesh is shared between computation nodes, the If mesh is shared between computation nodes, the parallelization is not trivial, not as easy to gridifyparallelization is not trivial, not as easy to gridify

Direct aproachDirect aproach

IMPLEMENTATIONIMPLEMENTATION

Adaptation of IVISDEP TO INT.EU.GRIDAdaptation of IVISDEP TO INT.EU.GRID

Use case:Opengl+GLUT+MPI+i2glogin+ Use case:Opengl+GLUT+MPI+i2glogin+ Gvid+Migrating DesktopGvid+Migrating Desktop

IVISDEP running on int2grid.IVISDEP running on int2grid.

Different instances of IVISDEP running on Different instances of IVISDEP running on the grid and visualized using MD.the grid and visualized using MD.

Direct ApproachDirect Approach

Not implemented, only proposed.Not implemented, only proposed. Viability:Viability:

Computation: O(Nv2 Np2)

Transmision (Nv1/3 +1)2 Np

Direct Approach: An exampleDirect Approach: An example

Node A Node B

100 Hex (100x27 modes)

(1 sec/iteration)

100 modes * 16 data * 8 bytes/data =

12.8 kbytes/surface

100 Hex (100x27 modes)

(1 sec/iteration)

12.8kbytes *100 Hex=1.2 MB/s

Node 2 Node 3 Node 4 Node 5 Node 6 Node 7Node 1

Total bandwidth/node=2.4MB/s full duplex

Data management and Data management and visualization on the gridvisualization on the grid

Analyze and visualize the big amount of data Analyze and visualize the big amount of data generated (more than 100GB per simulation):generated (more than 100GB per simulation):• If looking for something it is not easy what to look at. If looking for something it is not easy what to look at. • So, ideally, the more you can see the more you can find.So, ideally, the more you can see the more you can find.

Data distributed around the SE on the grid.Data distributed around the SE on the grid.

Could be possible for the user to:Could be possible for the user to:• Interact with the remotelly stored dataInteract with the remotelly stored data• Analyze data in real time on the gridAnalyze data in real time on the grid• Visualize a complex graphical representation generated on Visualize a complex graphical representation generated on

the gridthe grid

Proposed modelProposed model

ConclusionsConclusions

Grid has demostrated to be useful for the Grid has demostrated to be useful for the Fusion community through developed Fusion community through developed INT.EU.GRID applications (IVISDEP)INT.EU.GRID applications (IVISDEP)

Next applications seem to be more parallel Next applications seem to be more parallel demanding codes. Here we propose onedemanding codes. Here we propose one

We propose too an interactive data We propose too an interactive data visualization and management application visualization and management application for scientific results analysisfor scientific results analysis

Thank you!Thank you!

Questions?Questions?