COMMUNICATIONS IN APPLIED NUMERICAL METHODS, Vol.

3, 315-317 (1987)

A CHEMICALLY ENHANCED OIL RECOVERY

SIMULATION BENCHMARK
KAMY SEPEHRNOORI, CHOWDHURY K . MAMUN A N D GARY A . POPE
Department of Petroleum Engineering, The University of Texas at Austin, Texas 78712, U.S.A.

SUMMARY
The results of a benchmark on 15 computers for a chemical flooding simulator are reported in this paper.
Comparisons between a single processor and multiprocessors are made.

INTRODUCTION

The purpose of this paper is to report the results of a benchmark for a compositional reservoir
simulator of chemically enhanced oil recovery processes (UTCHEM). ' The same simulation was
made on 15 computers using a single processor and on two computers using multiprocessors.
These 15 computers represent a wide cross-section of scientific computers. This benchmark is
interesting because it represents the solution to a realistic problem of great interest to the petroleum
industry and is significantly different from other published benchmarks.
By a reservoir simulator we mean a numerical model of the permeable medium in which fluid
displacement is occurring. In general, this requires the numerical solution to a set of coupled,
nonlinear partial differential equations describing complex physical processes in a complex three-
dimensional geometric domain. For the chemical process in particular, these processes involve
multiphase flow of a large number of components which interact with the porous medium and in
other ways lead to extreme nonlinearities in the partial differential equations. Thus, the solution
to this type of problem is difficult even compared to other reservoir simulators, such as those used
to model waterflooding.2 The partial differential equations are discretized by using a finite
difference or finite element method, which in either case yields a system of algebraic equations.
This system of equations must be solved at each time level by means of a solution algorithm. The
particular simulator and reservoir displacement problem used for this benchmark are briefly
described below.

CHEMICAL FLOODING SIMULATOR

UTCHEM treats the transport of chemical species in three-dimensional multiphase flow through
permeable media subject to viscous, gravity, capillary and dispersion forces. The chemical species
include surfactants, polymer, various electrolytes, water and oil. The phases formed by these
components are brine, oil and microemulsion. Major physical phenomena modelled in UTCHEM
include dispersion, adsorption, interfacial tension, relative permeability, capillary trapping of
residual oil, brine and microemulsion, phase behaviour, viscosity, capillary pressure, inaccessible
pore volume and permeability reduction by the polymer. The phase behaviour is particularly
complex and important for this process.
The mathematical formulation for this model includes a pressure equation and ( n - 1 ) mass
conservation equations for the n chemical species. These differential equations are discretized
using a second-order, finite difference approximation for the spatial derivatives and a forward-
difference approximation for the time derivatives. The pressure equation is solved implicitly using
explicit concentration-dependent terms. The species conservation equations are solved explicitly.
Because of the extreme nonlinearity of the model equations, the allowable time-step size is

0748-8025/87/040315-O3$05.OO Received January 1987

@ 1987 by John Wiley & Sons, Ltd.
3 16 K . SEPEHRNOORI, C. K . MAMUN A N D G . A. POPE

extremely small, so the time integrations involve a very large number of floating point arithmetic
operations. This large computational requirement is in this respect characteristic of many large-
scale reservoir simulation problems.

BENCHMARK

This study was carried out in two different operational modes. First, the code did not incorporate
any restructuring to exploit vector or multiprocessing capabilities of the computers under consider-
ation. No assembly language coding was used, nor was any effort made to use special hardware
features. However, full use of any compiler optimization options was made. We refer to this class
of run as ‘unmodified’. The second mode of operation incorporated coding changes in order to
take advantage of the multiprocessor architecture of the VAX 8800 and Alliant FW8. The only
vectorization involved was that done by the compiler on the Alliant. The results using this
‘modified’ code should be considered as exploratory in nature, and we believe better performance
can be obtained upon further code modification. All the computations were performed using 64-
bit arithmetic, with the exception of the CDC CYBER 170/750, which used 60-bit arithmetic.
The particular problem simulated in this study was a chemical flood of a three-dimensional,
quarter five-spot, well pattern with an area of 5-47 acres. The simulation was performed for 50
days with a time-step size of 0.5565 days. The reservoir was discretized using a mesh of NX= 11,
NY=11, N Z = 2 for the x , y and z directions, respectively.
The results are given in Tables I and 11. Table I shows the computational time for the unmodified
code, and Table I1 shows the wall clock time for the modified code. The computational times for
the 15 computers shown in Table I ranged from 49 to 1228 sec, and in all these cases only one

Table I. Computation time (CPU) for the

unmodified code

Computer Time
(set)
__-
VAX 11/780 1228
VAX 11/785 812
CDC CYBER 1701750 591
ELXSI 341
VAX 8600 337
FPS 164 334
CONVEX C-1 326
Alliant FW8 (1 CE) 290
VAX 8650 234
VAX 8800 223
CONVEX C-lKP 204
FPS 264120 150
FPS 264 88
CDC CYBER 205 78
CRAY X-MP 49

Table 11. Wall clock time for the modified

code

Computer

Alliant FW8 (2 CEs)t 20 1

Alliant W 8 (4 CEs) 144
Alliant FW8 (8 CEs) 115
VAX 8800 (2 processors) 135

t CEs, computational elements.

 AN OIL RECOVERY SIMULATION 3 17

processor was used. It should be noted that the relative performance of these computers may vary
using other types of codes with structures different from that of this reservoir simulator. As shown
in Table 11, the wall clock time for the modified code ranged from 115 to 201 sec. The time for
the Alliant decreased from 290 to 115 sec as the number of processors increased from one to
eight. The time for the VAX 8800 decreased from 223 to 135 sec when two processors were used.
We have also explored the impact of compiler vectorization of this code. The simulator was run
with the vectorization option of the compiler turned on and off for the CRAY X-MP and the
CONVEX C-l/XP. There was hardly any difference in either case. For the CRAY, the times were
48.8 sec with this option turned on and 51-5sec with this option turned off, and for the CONVEX
these times were 204 and 206 sec, respectively. This shows that this code does not lend itself to
compiler vectorization.
In a previous study,3 we reported the results we obtained when we optimized the UTCHEM
code to take advantage of vector processing on a CYBER 205. Depending on the mesh size, the
speed-up factor ranged from 12 to 28. The benchmark time of 78 sec given in Table I for the
CYBER 205 decreased to 6.5 sec. More recently, we obtained similar results for the speed-up
factor when we vectorized the code for the CRAY X-MP. This shows that both multiprocessing
and vectorization have significant potential when the code is modified to take advantage of these
features.
ACKNOWLEDGEMENTS

This study was made possible in part by the financial support of the various participating companies
of the Center for Enhanced Oil and Gas Recovery Research at The University of Texas at Austin.
We would also like to express our appreciation to each of the computer companies cited for their
courtesy and assistance.
REFERENCES
1. A. Datta Gupta, G . A. Pope, K. Sepehrnoori and R. L. Thrasher, ‘A symmetric positive definite formulation of a
3-D micellar/polymer simulator’, 1. SPE Res. Eng., 1(6), 622-632 (1986).
2. K. Aziz and A . Settari, Petroleum Reservoir Simulation, Applied Science, New York, 1979.
3. A . Datta Gupta, B. Mossberg, G. A. Pope and K. Sepehrnoori, ‘Application of vector processors to chemical enhanced
oil recovery simulation’, Comrnun. appl. numer. methods, 2(3), 297-303 (1986).