Towards a More Effective Parallel Reservoir Simulator

Abstract

In 1989 it has been reported how an operational black/volatile oil reservoir simulator of IMPEC type has been successfully parallelised for a medium-scale (10-100 processors) local memory MIMD architecture1. The present paper reviews the principles underlying the parallelisation and describes the further development of the parallel simulator. Some attention is given to the linear solver which is of conjugate gradient type, with a local, block-wise preconditioner. A novel acceleration method, CGSTAB, for the solution of non-symmetric linear systems is described. The parallel simulator is now being evaluated in an operational environment. It has been installed on a 90 processor, 5500,000 Meiko M40 that should support two simultaneous job streams, each running at one quarter to one half (depending on the input) of the speed of a 1 processor Cray-XMP. So, at present, parallel computing leads to a substantial reduction of computing cost. In the very near future however, parallel computing machinery of the same family but based on the latest hardware, will beat traditional sequential computers on absolute performance too.

Development of a Scalable Thermal Reservoir Simulator on Distributed-Memory Parallel Computers

“…SAGD process with 756 well pairs is simulated. The grid has a dimension of 60 × 220 × 85 and size of 20 f t × 10 f t × 1 f t. All wells are horizontal wells along x direction, if the index of y direction of a grid block equals to 3,7,11,15,19,23,27,31,35,39,43,47,51,55,59,63,67,71,75,79,83,87,91,95,99,103,107,111,115,119,123,127,131,135,139,143,147,151,155,159,163,167,171,175,179,183,187,191,195,199,203,207,211, or 215, and the index of z direction equals to 4,10,16,…”

“…Many preconditioners have been proposed to accelerate the solution of linear systems, such as constrained pressure residual (CPR) methods [10,11]; multiple level preconditioners [12]; multi-stage methods [13]; CPR-FP, CPR-FPF, and CPR-FFPF methods [14]; and FASP (fast auxiliary space preconditioners) [15,16]. Parallel computers have more memory and better performance, which provide excellent approaches to accelerate reservoir simulations [17][18][19][20][21][22][23][24][25][26][27][28]. Wang [29,30] implemented a fully implicit equation-of-state compositional simulator for distributed-memory parallel computers, and large-scale reservoir models were simulated [31].…”

A Scalable Thermal Reservoir Simulator for Giant Models on Parallel Computers

“…SAGD process with 756 well pairs is simulated. The grid has a dimension of 60 × 220 × 85 and size of 20f t × 10f t × 1f t. All wells are horizontal wells along x direction, if the index of y direction of a grid block equals to 3,7,11,15,19,23,27,31,35,39,43,47,51,55,59,63,67,71,75,79,83,87,91,95,99,103,107,111,115,119,123,127,131,135,139,143,147,151,155,159,163,167,171,175,179,183,187,191,195,199,203,207,211, or 215, and the index of z direction equals to 4,10,16,…”

“…In the early stage, vectorization techniques in shared-memory machines was widely applied though it didn't scale very well [9,12]. Meijerink [23] developed a black oil simulator using the IMPES method and implemented on a local-memory MIMD computer. Chien [7] applied domain decomposition and MPI on an IBM SP-2 parallel computer.…”