Parallelization of a furnace simulation code for multi-core machines

Abstract

Şişecam uses Fortran programs that implement mathematical models for designing glass furnaces and for planning optimized operational criteria. The mathematical model for glass furnaces employs revised version of Patankar and Spalding's Semi- Implicit Method for Pressure-Linked Equations method called SIMPLER. The main objective of this thesis is to parallelize the sequential modelling programs using Open Multi-Processing (OpenMP) for shared-memory multi-core architectures. The sequential program uses Tri-Diagonal Matrix Algorithm (TDMA) for solving the resulting linear systems of equations. Sequential code is analyzed by pro ling in order to locate sections that have long running times. Additionally, another linear system solver, Stone's Strongly Implicit Procedure (SIP), is integrated to the program. Radiation calculations and linear system solvers which are the most computational intensive parts of the program are parallelized. The performance of the parallelized code is compared that of the sequential code. Results obtained on an Intel 8-core Xeon system show that speed-ups of roughly four times are possible.