A large-scale compositional simulation (;This dissertation presents memory management, programming and computational techniques that fully exploit the capabilities of parallel supercomputers for a large-scale compositional simulation. Examples of such techniques are the allocation of variables to dynamic memory, the implementation of the Young formulation and the use of the newly developed Sequential Staging of Tasks (SST) technique. The Sequential Staging of Tasks is a novel technique that can take full advantage of parallel processing to speed-up the solution of a large linear system.;The SST technique is a state-of-the-art parallel algorithm which has a number of advantages when compared to conventional domain decomposition techniques: (1) SST can use any number of parallel processors without a need for recoding, (2) SST automatically keeps good load balance among the processors, (3) SST significantly reduces paging when the problem size exceeds the processor storage, (4) SST preserves vectorization speed-up, (5) SST makes use of readily available iterative solvers, (6) SST is much easier to implement, and (7) SST yields substantially higher parallel speed-up.;Proper memory management allows for compositional simulations as large as 200,000 cells. Vectorization provides approximately a seven times reduction in CPU time requirements. The SST technique yields a 5.6 speed-up with six parallel processors. The compositional simulator developed with the techniques presented in this work provides a turn-around time over one hundred times faster than on a conventional serial computer.
展开▼