Analysis and Visualization of Stellarator Plasma Confinement Devices
Benefits of Parallelization depend both on the scientific model used as well as algorithm development.
Description of the Problem:
Axially symmetric (2D) tokamak magnetic surface with field lines
Stellarator optimization loop determines the outer flux surface shape. Coils which produce this shape are then derived:
PPT Slide
Three classes of orbits are important in stellarators: - particle class determined by: angle of velocity vector with magnetic field and particle energy - complicated orbit phase space requires simulations with large numbers of particles
Our model follows non-interacting particle trajectories through 4D phase space colliding with a static background plasma modeled with a 2D velocity space Monte Carlo diffusion operator:
A short list of reference material on MPI
Structure of DELTA5D Monte Carlo Code (MPI mostly involved at beginning and end):
MPI Startup
Processor Specialization (i.e., insure that each PE is following it’s own unique set of orbits)
Sum ensemble average information (for each time step) using MPI_REDUCE:
Stack particle time history information using MPI_GATHER:
Stellarator/tokamak Monte Carlo model on the T3E shows linear scaling:
The parallel T3E particle code has allowed us to examine problems which would otherwise be unfeasible
I have recently started merging two components of the ACTS Toolkit with my stellarator Monte Carlo code: CUMULVS and PVODE
CUMULVS
Information Sources on CUMULVS and PVM
CUMULVS Installation Issues
First, start up PVM on the T3E, spawn the executable, then add local workstation to the parallel virtual machine:
Now going over to the local workstation (FEDU55) one can connect a viewer application to the virtual machine and interact with the T3E job:
Other useful PVM diagnostic information:
PVODE Information Sources:
PVODE Installation Issues
Use of PVODE in Fortran codes:
Use of PVODE in Fortran codes
Conclusions
Email: TMDeBoni@LBL.GOV
Download presentation source