The EAGE-sponsored workshop on open source E&P software was a reprise of an event held six years earlier. BP’s Joe Dellinger set the scene with a recap of the open source movement in general and of Unix-based seismic systems including SEPLib, Seismic Un*x (SU) from the Colorado School of Mines and Amoco’s USP. The ‘reproducible’ movement originated at Stanford where the inclusion of reproducible code is now mandatory. The 2000s saw a wave of oil company mergers and a software shake out—it was easy to leave with software like FreeUSP, FreeDDS or with Amoco’s synthetic seismic data sets. The last decade saw the development of CPSeis from ConocoPhillips and JavaSeis. Madagascar was also released and reproducibility made easier while Python based processing systems proliferated. DGB’s Open dTect got a mention as an open source commercial success. So what’s in store for the 2010s? GPU-based software is a flash back to earlier days of the array processor. Big data and big graphics are important trends. Companies will ‘cover walls of rooms with high resolution screens.’ And maybe open source will see industrial strength and easy to install software that has so far been lacking.
Columbia’s Victoria Stodden described the central role of geophysics in reproducible research—thanks to John Claerbout’s influence in research and on US policy. The reproducible movement came about in reponse to a credibility crisis in computational science in the last century. In the June 1996 Journal of the American Statistical Association, nine out of 20 papers were computational and none included code. By 2011, 29 out of 29 articles were computational and 21% made their code publicly available. An article about computational science is not the scholarship itself, it is just an advert for the scholarship. The scholarship is the complete software development environment and the code. Why is all this important? On the one hand, computation is now a branch of science in its own right. On the other hand there is the ‘ubiquity of error’ and a lot of loose thinking that tends to generate ‘breezy demos,’ not reliable knowledge. Intellectual property is an issue for open source. But much thought has been devoted to fixing this with notably the Apache Foundation’s licensing, Creative Commons and Stoddens own work.
Bjorn Olofsson presented SeaSeis—his open source sequential pre stack batch seismic processing system. SeaSeis is an Ascii control file based workhorse with 80 modules and a 2D viewer. The system is based on the idea that ‘simple tasks should be simple to run.’
Chuck Mosher (ConocoPhillips) spoke on parallel I/O and computing in JavaSeis. Originally developed by Arco, JavaSeis is used by ConocoPhillips and Halliburton. JavaSeis is run from the Eclipse IDE and offers Matlab integration.
Chris Jones explained that BG Group uses and supports open source seismic software because it gives BG’s seismologists the ability to do what they want, to experiment with algorithms in a scalable, flexible environment. BG deploys an ‘agile’ approach atop a huge Fortran 77 code base. Open dTect’s attribute engine, Lustre and Slurm job management are used, as are visual programming tools like Scratch, MIT’s AppInventor and the ubiquitous Perl. Yanghua Wang’s Multichannel matching pursuit algorithm has been ported to Open dTect. Other use cases include Baysian classification of seismic facies and depth sensitivity analysis. BG uses Matlab (grads are more familiar with Matlab than Python) and for performance—Matlab can also be run on the cluster. BG is interested in running Seismic Un*x in Hadoop as it is ‘especially good for parallelization.’ I/O remains a sticking point—JavaSeis is ‘half way there.’ Service providers should provide open source solutions and more reusable code a la Matlab. More from the Workshop website.
© Oil IT Journal - all rights reserved.