document

Abstract

The Connection Machine, which uses over 65,000 processors in parallel, is a SIMD (Single Instruction Multiple Data) computer capable of computing large reservoir simulation models efficiently. Features of this computing system are ideally suited to finite-difference solutions where equivalent operations are performed on different data in the grid performed on different data in the grid blocks. The Connection Machine hardware and software are described, the application to reservoir simulation discussed, the code for a simple example problem provided, and timing results for this problem tabulated.

Introduction

New computing architectures have the potential for allowing problems to be potential for allowing problems to be computed which were previously not possible or practical. The greater computing capability provided by these machines can be utilized in calculating finite-difference reservoir simulation models containing many cells. Computers capable of computing these larger more detailed models at greater calculation speeds will enable greater resolution of important geologic features to be incorporated in reservoir models and will allow reservoir energies and production mechanisms to be considered over an entire reservoir. Including such features in models will allow more accurate modeling and predictions. predictions. Parallel-computing systems can be distinguished by the number of processors used and their arrangement and also by the structure of their memories. Shared-memory systems use a small number of powerful processors which access a common memory. processors which access a common memory. Distributed memory or local memory computing systems are constructed with each processor having its own individual memory processor having its own individual memory separate from the memory associated with other processors. In this type of system data is exchanged between processors by message passing. Hybrid systems have features found in both the other two types of systems. Computing systems can also be distinguished by the way instructions are performed by the processors. If all performed by the processors. If all processors use the same set of instructions processors use the same set of instructions and execute this set simultaneously on different data the computing system is said to be SIMD (Single Instruction Multiple Data). When each processor can perform a different set of instructions on a different data set the system is MIMD (Multiple Instruction Multiple Data). The Connection Machine is a SIMD computing system with a distributed memory. All processors execute the same instruction simultaneously using data in their individual memories. Because of the fine-grained structure of this system one processor can be associated with one or processor can be associated with one or more data elements in a problem. For finite-difference applications this means that one or more cells or computational elements can be associated with each processor. Each processor performs the processor. Each processor performs the calculations for the cell or computational unit whose data is stored in its memory. This allows the parallelism found in many problems to be exploited. problems to be exploited. P. 7