Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.654203
Title: Parallel processing methods applied to two and three dimensional geo-electromagnetic induction modelling
Author: MacDonald, Kenneth J.
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1996
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Two existing finite difference algorithms for solving the forward modelling problem of geoelectromagnetic induction have been recoded to take advantage of high performance massively parallel SIMD (single instruction multiple data) computer architectures. Poll's solves the two scalar polarised fields in the two dimensional (2D) problem, and the other from Pu solves for all three components of the magnetic field in three dimensional (3D) structures. Both models apply integral boundary conditions at the top and bottom of the grid to limit total mesh size. The 3D model introduces a thin sheet at the top of the model to describe near surface features. An efficient data parallel algorithm ensures the evaluation of the integrals maintains a high ratio of processor utilisation on the parallel hardware. Data parallel versions of the point Jacobian, Gauss-Seidel and successive overrelaxation iterative solvers have been developed. The latter two require two level black-white ordering, which to equalise the processor load balance, has been implemented in both a horizontally banded and chequer boarded remapping of grid nodes. The 2D model was also developed to form a task farm, whereby the solution for each period is performed on one of a cluster of workstations. These solutions are independent of each other, so are executed simultaneously on however many workstations are available at the time. Modern workstations, coupled with the original 2D Gauss-Jordan solver, are faster than the SIMD computers for all but the largest grid sizes. However, the 3D code certainly benefited from the parallel processing for any but the smallest models. A new automatic meshing algorithm, which stretches a predefined number of grid points over the conductivity structure, has also been developed. In part, this was to control the mesh sizes and hence load balancing on the SIMD computers, but investigations into grid spacing for 2D models show that severely restricting the number of grid points results in a much faster estimated solution.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.654203  DOI: Not available
Share: