Large-scale parallel geophysical algorithms in Java: A feasibility study

摘要

The implementation of some seismic processing methods in procedural programming languages like Fortran (which are widely used in the petroleum industry) has severe disadvantages: 1. Procedural programming languages usually lack "software reusability," which makes it difficult to maintain and extend the source code. The Stanford Exploration Project tried to use object-oriented C++ to develop a software framework for easy programming of seismic processing and inversion operators. Its efforts produced reusable operators, but it had problems when working on realistic prestack seismic data sets. The added flexibility was countered by long computation time. Moreover, it is not possible to express parallelism of execution in a standard and portable way. 2. Message passing libraries (often used to implement seismic methods on parallel distributed memory computers) force the programmer to deal with the geophysical problem and also to tackle the problems of parallel programming (e.g., load balancing and data locality). The resulting code is often error prone and cannot easily be ported.