Macro-parallelisation for controlled source electromagnetic applications

© 2015 Elsevier B.V. Many geophysical computational problems can be referred to as "embarrassingly parallel". Parallel computing utilises linked CPU cores to solve computational problems. We create a "macro" parallelisation method that rapidly recovers solutions to large scale el...

Full description

Bibliographic Details
Main Authors: Pethick, Andrew, Harris, Brett
Format: Journal Article
Published: 2016
Online Access:http://hdl.handle.net/20.500.11937/13335
Description
Summary:© 2015 Elsevier B.V. Many geophysical computational problems can be referred to as "embarrassingly parallel". Parallel computing utilises linked CPU cores to solve computational problems. We create a "macro" parallelisation method that rapidly recovers solutions to large scale electromagnetic forward and inverse modelling problems. The method involves software operating above a generic electromagnetic data structure. Two examples are provided. The first example quantifies the reduction in computational time where macro-parallelisation is applied to forward modelling of data generated during synthetic marine controlled source electromagnetic surveys. In the second numerical experiment we apply macro-parallelisation to recover the subsurface conductivity distribution from a large airborne transient electromagnetic survey spanning more than 2000km2. The computation time for inverting 98 thousand soundings with a serial batch approach on an i7 with a single thread was 65h. Computational time from inverting 98 thousand soundings on a single thread of a standard i7 processor was 65h. A 1700 times improvement in computation time was achieved through macro-parallelisation across just 350 cores of a Cray XC30 Supercomputer. Inversion of data for the full AEM survey took just 135s. Parallel computing is rapidly becoming an essential for geophysicists. We provide description, sequence diagrams, pseudo-code and examples to illustrate its implementation. In summary we present applied parallelisation for the masses.