120 likes | 264 Views
Data Optimization for Reflectometer Simulation. Marc Osherson Science Education Summer Intern 9/1/2005. FWR. A 2D simulation of a full wave reflectometer for which Eliot Feibush has been developing a GUI. Client-server architecture. GUI can run on a user’s laptop.
E N D
Data Optimization for Reflectometer Simulation Marc Osherson Science Education Summer Intern 9/1/2005
FWR • A 2D simulation of a full wave reflectometer for which Eliot Feibush has been developing a GUI. • Client-server architecture. • GUI can run on a user’s laptop. • Input and Output netCDF files are on the PPPL cluster. • Files had to be transferred to the laptop, and processed there.
Multi – Tier Software Architecture PPPL Compute Cluster TRANSP genxpl trxpl User’s PC, Mac, Linux Java Servlet Java Client GUI FWR Input netCDF Files MDSplus Server TRANSP Runs netCDFcon FWR Solution Correlation Graphs FWR Simulation
Problem • Input and output files became prohibitively large for laptops, e.g. 100MB output files. • The files would take too long to transfer and process. • The Java heap memory was too small and could not handle such files.
Approach • Server side post-processing of input and output files. • Reduce simulation data to just the displayable image data. • Arrays of doubles converted to byte data. • Provide a means for recovering the actual data values.
netCDFcon (C program) • Converts the plasma input file into efficient image data. • Computes the three cutoff images and scales them into image data. • Converts the full wave solution file to image data. • Runs on the PPPL compute server, much faster than running Java on a laptop. • Roughly 100:1 speed up. • Converts a 100MB solution file to 1MB. • Significantly reduces file transfer time. • Convert once, display many times.
netCDFcon • Linux command line program. • Command line arguments: • Input file • Output file • Type of processing • List of variables, e.g. ne te bb • Run by the servlet in a script file.
FWR Demo Questions ?
Thank you • To the entire science education staff for making this possible. • To Eliot Feibush for all the help he’s given me. • To PPPL for this opportunity.