Grid Cluster in the Office: High-Performance Computing for Reservoir Management

Abstract

Optimal reservoir management usually requires evaluation of a large number of various development scenarios. This applies to a choice of well placement and settings configuration, waterflooding strategy and history matching process. Reservoir models are commonly used to obtain field production forecast depending on the development plan. Generally, a specialized external high-performance computational environment is used to accomplish the aforementioned multivariate calculations, while personal workstations stay idle most of the time. We propose an alternative way of organizing HPC environment using available computational resources. The method is based on integration of several personal workstations into a so-called grid cluster using local network, which allows managing the whole system from any connected node. Each computational node is provided with a flexible timetable to ensure that distributed calculations do not interfere with daily work. The solution does not require additional capital investments and is easy to implement in the office. Although initially designed for hydrodynamic simulations, the system can be used for any time-consuming multivariate task. Proposed method has been applied to several optimization cases of real fields. Grid cluster consisting of fifty nodes with estimated peak performance of 60 TFLOPs was used to find optimal development plan for a field in Western Siberia, allowing to reduce computational time from several months to one weekend. The system prooved linear speedup depending on the number of involved workstations along with stability under conditions when each node may connect or disconnect at any time.