Keywords: high-performance computing, service-oriented architecture, geophysical well-logs modeling inversion interpretation, parallelization framework, cluster multi-core computing.

Three-dimensional modeling required for interpretation of geophysical well-logging data in non-vertical wells is hardly feasible on a single-processor machine and is rarely used, because the simulation time is prohibitively long. However, most solvers are embarrassingly parallelizable, as responses at each tool position are independent of each other. A distinctive objective of our research is to facilitate solving the inverse problems "just in time" for a driller, a reservoir engineer, a log analyst, etc.We developed a service-oriented HPC framework for interpretation of well-logs, putting physics-based modeling in the oilfield practitioners' hands. At the heart of the system are parallelized simulation algorithms. Exposed through the Web Services API, these codes become a network-accessible library of core simulation routines available to any consumer application on the network.

The foundation of the system is Grid Services Lite, a cross-platform, service-oriented, easily deployable, highly efficient parallelization infrastructure, capable of O(10e3) task throughput on O(10e5) worker threads, with well-optimized load balancing between multiple simultaneous users. The system performance was evaluated using the methodology described in [1].

Parallelized forward modeling and inversion simulators exposed as a Web Service is a key innovation in the exploration and production industry, being an enabling technology for a spectrum of business-critical workflows [2]. The Log Modeling Service is a GSL client responsible for executing forward modeling and inversion tasks. The library of simulators includes optimized forward modeling codes of electromagnetic, sonic, and nuclear tools, as well as parametric inversion engines. GSL demonstrates the same efficiency with these log modeling codes as on synthetic benchmarks, as well as well-optimized load balancing between multiple simultaneous users.

Our future effort will be directed at enabling a significantly larger number of node connections to the GSL server through the use of scatterer-gatherer gateways; optimizing memory requirements in the case of very large individual jobs; and adaptive optimization of the atomic unit of work size.

1

I. Raicu, Y. Zhao, C. Dumitrescu, I. Foster, M. Wilde, "Falkon: a Fast and Light-weight tasK executiON framework", IEEE/ACM SuperComputing, 2007. doi:10.1145/1362622.1362680

2

M. Wiig, E. Berg, J.M. Kjaerefjord, "Geosteering Using New Directional Electromagnetic Measurements and 3D RSS in a North Sea Well", paper 95725 presented at the 2005 SPE Annual Technical Conference and Exhibition, Dallas, TX, 9-12 October 2005.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £72 +P&P)