110 likes | 117 Views
This article provides an update on the Toric development for Transp, including version information, code status, and integration with Transp. It also discusses available tools for data analysis and visualization.
E N D
Toric at JET –Status & Tools Work in Progress – See http://www.jet.efda.org/expert/transp/Toric/index.htm for latest information Background • RF models available in Transp are obsolete, unsupported or both • Reliable RF code will be needed to analyse ILA pulses • We agreed with Doug McCune that JET would implement latest version of Toric in Transp • Project aims are well defined, but no end user identified, as yet.. TF-H 20-May-2008
Toric Versions Toric is available in different versions & modes of operation – • All code compiled with Lahey Fortran ( lf95 ) • Parallel version runs under OpenMPI • Requires BLAS, LaPack & NETcdf libraries • Initial attempt to use PGI Fortran abandoned, LaPack failed acceptance tests TF-H 20-May-2008
Status of the Code Toric was converted to use SI units, between versions 4.2 & 5. The power deposition & driven current calculations were found to be inconsistent, when processing the supplied test data with both versions. • Normalisation of T5 has been corrected to reproduce the T4.2 current drive calculation.However, the possibility remains that T5 was correct & T4.2 was wrong • Power balance ( written to log file ) differed by x2, for the test shot ( fixed ) • Field amplitudes written by transp interface code differ by ~1/20(E), 10-7 (B) between versions, need to change normalisation to compensate for this ( or restore previous values ) Toric5 / Transp interface code requires further verification,before being accepted for production runs TF-H 20-May-2008
Initial Results from Transp / Toric 5 Comparison with Pion( dashed lines ) for 69409 – Minority He heating : TF-H 20-May-2008
Parallel Toric • Timing for a single antenna, single timepoint, varying Nmod and number of processors – • Using 16 processors gives a modest improvement in elapsed time, but doesn’t compensate for increased Nmod ( poloidal mode number) TF-H 20-May-2008
Toric and Transp Integration of parallel Toric code with Transp is non-trivial – • Transp itself is currently a ( large ) serial code • Other modules ( notably nubeam MonteCarlo ) take a significant time • Cant leave 15 processors idle while one runs nubeam.. Options – • Clever job scheduling, share processors between different transp runs • Run nubeam & Toric in parallel ( on same time slice ) ? TF-H 20-May-2008
Data - Analysis & Display Toric Input ASCII file specifying equilibrium – Nx, Tx vs. Phi Output NETcdf file / antenna, Power deposition + Wave fields Transp Outputs NETcdf file , ~ 1500 plots/run + 10 more NETcdf files related to FP /QLO may be saved at selected timepoints. Tools IDL - JETdsp will read Transp NETcdf Toric provides some IDL utilities Python - matplotlib.py recommended by MIT Toric.py developed to plot profiles & power deposition MATLAB - NO – unsupported across future Linux upgrades, cant afford to update the licenses TF-H 20-May-2008
IDL tools Toric IDL utilities – • fixed set of plots for a single run • Saved as .ps / .eps only • Limited control over layout JETdsp – • Cant read NETcdffiles from arbitrary locations • Need to work with complex existing data structures to make any changes “in my humble opinion its too expensive, too slow compared to compiled languages like C++, too much of a memory hog and worse still, too inelegant a language compared to modern interpreted languages like Python and Ruby.” TF-H 20-May-2008
Python / Matplotlib.py • Large number of modules available on JET analysis cluster • Example codes run if cut & pasted from web pages • Reads NETcdf files ( need ScientificPython module ) – which also understands Fortran format spec • Classes, inheritance etc better integrated than e.g. Perl. • FREE !! Toric.py • Plots profiles, power deposition from Toric outputs • PPF system interface is available • GUI under construction - TF-H 20-May-2008
Toric.py - Graphics TF-H 20-May-2008
Questions Toric development would benefit from some RF physics expertise – • List of shots for benchmarking, & output from other codes for comparison • What value ( or values ) of Phi should be used ? • How large does the poloidal mode number need to be ? • Do we need to run all 4 of the existing JET antenna at each timepoint ? • How should output be presented, what do you want to do with it ? Summary • Need to understand the code using existing RF system, ( & fix any bugs ) before adding ILA description • Available CPU / interconnect speed on JET cluster will restrict the accuracy of Transp RF runs, even using the parallel code. TF-H 20-May-2008