J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I....

8
J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski W. Wang

Transcript of J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I....

Page 1: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

J.-N. LeboeufV.K. Decyk

R.E. WaltzJ. Candy

W. Dorland Z. Lin

S. ParkerY. Chen

W.M. NevinsB.I. CohenA.M. DimitsD. Shumaker

W.W. LeeS. EthierJ. LewandowskiW. Wang

Page 2: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

The Plasma Microturbulence Project

Our Goal:Understanding plasma microturbulencethrough direct numerical simulation

• Plasma microturbulence is a critical issue to magnetic fusion program

– Controls energy confinement Determines size and cost of a

burning plasma experiment

• Direct numerical Simulation is the right tool to study microturbulence

– Far better diagnostics than those available on experiments

– Excellent resolution can be achieved on existing computers

Our Game plan:• Code Development

– Enhanced Fidelity

– Increased Efficiency

– Common data analysis & visualization system

• Code Validation

– Against each other’s codes

– Against experiment & theory

• Expanding our user community

– Web-based applications– Collaborations with theory

and experimental communities

Page 3: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Plasma Microturbulence is an Interesting Scientific Problem

Quasi-2-D turbulence in a 3-D toroidal geometry exhibits inverse cascades and other features of 2-D turbulent systems

Page 4: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Largest runs with ‘GTC’ code required 1 billion particles and 125 million grid points using 1024 processors on the IBM-SP at NERSC

Plasma Microturbulence SimulationsRequire State-of-the-Art Computers

Scaling of Plasma Microturbulence with System Size

Page 5: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Enhanced Code Fidelity

• Physics relevant for plasma confinement:

– Ion-scale Physics– Electron Dynamics– Turbulent Electric &

Magnetic Field– Realistic Geometry & Plasma

Size

• Accurate implementation demands:

– Excellent spatial resolution– High-order time integration– Advanced data-management

and visualization dealing with large data sets

QuickTime™ and a decompressor

are needed to see this picture.

Simulation Data from GYRO Code

Page 6: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Code Efficiency is a Critical Issue

Scaling of Particle-in-Cell Code (‘GTC’)Plasma Microturbulence

Project Codes:– Are largest users of computer

cycles within the Fusion Energy Sciences Program

– PMP codes Scale ~ Linearly with Number of Processors

– Production code efficiencies of 10-20% are achieved

Actively working with SciDAC Performance Evaluation Research Center to improve code efficiency

Page 7: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Advanced Visualization and Data Analysis Challenges

Terabytes of data are now generated at remote location (Data Management, Data Grid technologies)

Advanced visualization techniques needed to help identify key features in the data (Parallel Visualization)

Data must be efficiently analyzed to compute derived quantities

121 Milliongrid points

Temperature

Time

Particle in Cell Turbulence Simulation

Heat Potential

Page 8: J.-N. Leboeuf V.K. Decyk R.E. Waltz J. Candy W. Dorland Z. Lin S. Parker Y. Chen W.M. Nevins B.I. Cohen A.M. Dimits D. Shumaker W.W. Lee S. Ethier J. Lewandowski.

Data Management, Data Analysis & Visualization

• Terabytes/simulation is Data management issue

– Interactions with SciDAC Fusion Collaboratory

• Data must be analyzed– To validate codes– To gain insight

• Interactive data analysis– New insights from analyzing

data in new ways• Visualization of analyzed data

– Multi-dimensional data sets– Efficient means of

communicating between computers to people

Turbulent FluctuationSpectrum