The Petroleum Technology Transfer Council held a workshop1 in Houston last month to promote ‘reproducible’ geophysical computing with open software. Reproducible research is a new concept for communicating geoscience technology, both in the public domain and inside individual organizations. It was first proposed by Jon Claerbout in 1992 who observed that ‘electronic documents give reproducible research a new meaning.’ Computational experiments with geophysical data are captured in the form of transferable ‘recipes’ which can be shared and modified by users. The recipes are embedded in publications, typically ‘live’ hyperlinked documents. The main vector for reproducible geophysical computing are Clarebout’s Stanford Exploration Project (SEPlib) and Madagascar, but the PTTC meet was a venue for other open source geophysical software including Seismic Unix (SU), FreeUSP, DDS, JavaSeis, JTK, Pseis, OpendTect, CPseis and Madagascar.
Ioan Vlad described how Fusion Petroleum Technologies uses open-source software, noting two common misconceptions. Open-source software (OSS) is often perceived as incompatible with the activity of a commercial company. On the other hand, open-source packages do not currently provide a ‘complete solution’ for geophysical software.
Fusion uses OSS because it has fewer bugs than proprietary frameworks due to the greater number of coder users whose ‘many eyes’ find and fix bugs. Documentation is also usually better. Fusion has combined code from SU, SEPlib, Madagascar and CPSeis and integrated them with its GeoPRO Flow Builder preprocessing utility.
Richard Clarke (BP) revealed that FreeUSP/DDS (www.oilit.com/links/1107_31) usage amounts to 99% of usage of its research computing system, one of the world’s largest. The system is not used for production processing but its 450 teraflops bandwidth (40,000 CPUs) and 230 terabytes of physical memory demonstrate the scalability of the open source packages.
Karl Schleicher (University of Texas at Austin) noted the gap between a functioning research prototype and a tested program. His group is building a data library for testing open source seismic software. He also offered a detailed comparison of the different systems to conclude that Java and Python will slowly be adopted by industry and that although it is tricky to use, SU is the most mature package.
Alexander Popovici (Z-Terra) offered an in-depth comparison of SU, SEPlib and Madagascar to conclude that SU benefits from strong front-end time processing, although it lacks modern depth imaging algorithms. Madagascar has these along with wave-equation concepts, angle gather modules and more. SEPlib is now considered obsolete and will disappear when its functionality has been incorporated into Madagascar.
Bill Menger (Weinman Geoscience) presented, inter alia, ConocoPhillips’ GeoCraft (www.oilit.com/links/1107_32), a ‘lightweight’ framework for prototyping geoscience algorithms including AVO. GeoCraft is built on IBM’s ‘Eclipse’ rich client platform.
Yang Zang (Stanford) offered a cookbook and a movie showing how research can be rendered ‘reproducible.’ Key software tools are SEPlib, GNU Make, vplot, LaTex and Scons. More from www.oilit.com/links/1107_33.
1 Presentations available on www.oilit.com/links/1107_34.
© Oil IT Journal - all rights reserved.