Abstract
This paper presents the important considerations, programming and computational techniques for the development of a fully parallel and vectorized compositional reservoir simulator. A modified Young formulation is used for the flash calculations and a novel algorithm called Sequential Staging of Tasks (SST) that can take full advantage of parallel vector processing to speed up the solution of a large linear system is introduced. Compared to domain decomposition, SST has a number of noteworthy advantages illustrated with the results of computer experiments conducted on a parallel supercomputer with 6 vector processors. Memory-efficient techniques for out-of-core simulations are also discussed. The simulator developed with the techniques presented in this paper can handle compositional reservoir simulations as large as 200,000 grid blocks and components with reasonable turn-around times and can provide a 100 times speed-up over conventional serial computers.
