Pole Avenia’s Big data in geosciences conference

Total’s HPC-based ParaView. IFPen/Teradata, MapReduce in seismic inversion. CGG’s open source big geodata analytics. Ikats time series toolkit. Krakatoa deep learning for North Sea drillers.

Pole Avenia is geoscience-focused R&D incubator based in Pau, France. Speaking at a its conference on big data in geosciences*, Bruno Conche presented Total’s large scale data visualization engine, ParaView. PV has been implemented on Total’s Pangea supercomputer (page 4) and leverages spatial domain decomposition and parallel rendering on the server. The technique applies to visualization of both large seismic data cubes, giga-scale reservoir models and point cloud data.

Total is experimenting with compression technology from Hue (page 3), and on technology options from Nvidia (Index) and Intel (OSPray). ‘Google Map-like’ multi-scale gigagrid visualization has been co-developed with INT and Norway’s CMR Institute. NoMachine’s free remote desktop is also being used for desktop cloud visualization. A new paradigm is evolving of in-situ data viewing of data during processing, enabling QC and interaction without waiting for the final result. Here PV performs real time coprocessing inside the simulator.

IFPen researcher Hery Rakotoarisoa outlined work carried out in collaboration with Teradata on the application of MapReduce to seismic inversion. A Teradata Aster virtual machine demonstrated that MapReduce enables an in-database approach to the inversion of large seismic cubes.

CGG’s Guillaume Poulain also reported on the use of MapReduce in big geodata that benefits from its implicit parallelism. The open source Hadoop ecosystem is fault tolerant and capable of processing very large data volumes. MapReduce represents the ‘democratization of distributed calculation.’

Researchers from LIG, the Grenoble based IT research establishment, presented Ikats, an ‘innovative toolkit for analyzing time series.’ The system has been used to analyze a terabyte or so, nine months of flight data from four Airbus aircraft. R and Python code running on a single PC and using a subset of the data enabled Pearson correlation of multiple flight parameters.

Antoine Veillard introduced Krakatoa, his deep learning system for oil. Krakatoa has been used to investigate borehole quality and to ‘de-silo’ drilling and wireline data in a North Sea dataset, establishing machine learning-derived relationships between bad borehole caliper measurements and drilling parameters.

* Read the presentations here (mostly in French).

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.