Parallelizing an oil refining simulation: Numerical methods, implementations and experience

Abstract

One important process in oil refining is to separate the crude oil into various oil products. This process is called distillation. Engineers routinely perform a large computer simulation while designing a complex distillation column. This paper presents experience in parallelizing an oil refining simulation application that computes the composition of the various oil products in these distillation columns. Mathematical models for the simulation form large sparse nonlinear systems of equations. Triangular decompositions of these sparse nonlinear systems are fundamental numerical methods in this simulation. The paper shows the application of different approaches for carrying out this simulation in parallel. Parallelizations of the simulation program are introduced at three levels — direct, structured, and asynchronous. The approaches of this practical research provide insight into important issues of parallel computing for real-world applications. The paper includes parallel implementations on the Intel iPSC/860 and parallel computing results with comparisons and discussions.