The supercomputer Multi flow TRACE 7/200 was benchmarked with a compositional simulation model. For comparison, the model was also run on the CRAY X-MP/12, the CONVEX C-l XP, and the ALLIANT FX/8-8. The performance of the computer was judged in terms of CPU time and main memory requirements.

The Multiflow system uses the very long instruction word (VLIW) architecture. The instruction word can be as wide as 1,024 bits. It is packed with up to 28 primitive machine operations, which are executed in overlapped fashion. Thus, many different operations such as load, store, add, multiply, etc. may be executed simultaneously in one clock cycle.

The entrance level Multi flow system is the TRACE 7/200. It has a cycle time of 130 ns. Up to 7 operations may be executed on the 7/200 per instruction, resulting in 53 MIPs and 60 MFLOPs.

The object code for the Multi flow system is generated by the trace-scheduling compacting compiler. In addition to generating the compacted instructions for the VLIW, the computer predicts branch paths and prepares compensation codes to guarantee correct branching. The amount of compensation codes must be small in order not to degrade computing efficiency.

The reservoir model was an expanded version of the gas cycling model in the Third SPE Comparative Solution Project. It was derived by repeating the SPE model twice in the × direction and 4 times in the y direction. The dimensions of the new grid were thus 18 × 36 × 4. There were 8 injection wells and 8 production wells in the new grid.

The MULTIFLOOD simulator of Todd, Dietrich & Chase, Inc. was used to run the numerical model. The simulator is 3-dimensional, 3-phase, and compositional. It solves the system of partial differential equations by the finite difference technique and Newton-Rafson iteration. The system of linear equations is solved by Gaussian elimination with alternate diagonal ordering.

The scalar version of MULTIFLOOD was used in benchmarking. A ground rule was established that no changes were to be made to optimize the source code for each computer. However, the object code could be optimized by involving the appropriate compiler options or directives for vectorization on the CRAY and the CONVEX or for parallelization on the ALLIANT. On the TRACE, a fine tuning tool was used to assist the compiler in unrolling and in automatic inlining of the object code.

In evaluating the benchmark results, the simulator calculations are grouped into 4 categories. They are: physical properties evaluation, formation of finite difference coefficients, solution of linear equations, and miscellaneous calculations.

This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as presented, have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material, as presented, does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Papers presented at SPE meetings are subject to publication review by Editorial Committees of the Society of Petroleum Engineers. Permission to copy is restricted to an abstract of not more than 300 words. Illustrations may not be copied. The abstract should contain conspicuous acknowledgment of where and by whom the paper is presented. Write Publications Manager, SPE, P.O. Box 833836, Richardson, TX 75083-3836. Telex, 730989 SPEDAL.

This content is only available via PDF.