Master 2012 2013
Stages de la spécialité SAR
High performance computing applied to seismic imaging


Site :Intern Software Engineer . High Performance Computing aaplied to Seismic Imaging
Lieu :Meudon
Encadrant : Philippe Thierry
Dates :15/04/2013 au 30/09/2013
Rémunération :1200€ brut
Mots-clés : Parcours SAR autre qu’ATIAM, recherche


Warning: Illegal string offset 'id_auteur' in /dsk/www-master/html/2012/ecrire/public/assembler.php(625) : eval()'d code on line 3
Cliquer ici pour vous authentifier


Warning: Illegal string offset 'statut' in /dsk/www-master/html/2012/ecrire/public/assembler.php(625) : eval()'d code on line 1

Warning: Illegal string offset 'id_auteur' in /dsk/www-master/html/2012/ecrire/public/assembler.php(625) : eval()'d code on line 2

Description

In this position, you will be working with a small team of engineers researchers on software optimization and tuning

This internship is part of recently-launched and new technologies as the new multi cores Xeon E5 and many integrated cores Xeon Phi. The main goal will be to implement several new features in an existing seismic wave propagation and inversion code based on so called Full Waveform inversion. Such a code exhibits many different behaviors as huge floating points and memory bandwidth demands but also require a careful analysis and implementation of i/o and communication schemes.

Full seismic waveform inversion (FWI) is an iterative data fitting procedure, based on full waveform modeling, which directly extracts high resolution quantitative information on the subsurface from seismogram. Thanks to its accurate representation of the wave propagation physics it is considered as the "ultimate" seismic imaging technique allowing subsurface parameters reconstruction with a resolution equal to half the propagated wavelength. Most of the theoretical aspects of FWI have been worked out more than thirty years ago and its value demonstrated on few synthetic and field data examples. However, despite major research efforts to tackle the main challenges, FWI is still not a standard imaging technology in the oil industry due to several challenges, as computation cost for example.

The short term objectives of this project will be to port and evaluate the possible use of multi and many integrated cores to speed up the whole process. Such work will require the best use of programming model as MPI and OpenMP to take advantage of the full platform and the implementation of dynamic load balancing to avoid strong overhead. This will lead to the evaluation of high to low level programing models and their respective impact on development, portability, maintenance and performances. Careful profiling of current implementation as well as characterization and extrapolation of performances will be also needed to model performances of current and future FWI implementation on future hardware.

Qualifications

The successful candidate will be working toward a BS or MS in Computer Engineering or Computer Science with excellent communication skills, strong programming skills and a willingness to learn. Experience with Fortran , C , MPI and OpenMP is a must. Experience in optimized i/o and OS is a plus.