Steven Smith
Oxy USA Inc.
Tulsa
After decentralizing an exploration and production operation, a switch from mainframe computers to workstations decreased computing cost and increased capabilities for reservoir simulations In 1989, spurred by a major reorganization and the need to reduce reservoir simulation costs, Oxy USA Inc. began evaluating new technical computing strategies.
At that time, Oxy ran smaller simulations during the day on a dedicated IBM ES/3081 mainframe at the Tulsa headquarters. Larger runs were made overnight on a shared IBM ES/3090 mainframe in batch mode.
A team with members from management information systems (MIS) and the reservoir management group's simulation staff analyzed the cost of the mainframe usage.
Analysis showed that over $1 million was spent for mainframe services in 1988. But if the dedicated mainframe was removed, the available time on the shared mainframe during business hours would not be sufficient.
The study confirmed that each engineer needed unrestricted computer access for simulation during the work day. For this reason, the most productive and cost-effective approach would be to remove the dedicated mainframe and run daytime reservoir simulations on workstations, using the central, shared mainframe as a data server in a networked environment.
CHOOSING A WORKSTATION
To make the workstation selection, the team established a number of evaluation criteria.
- Each engineering workstation had to be a compute server with performance equal to or greater than that of an IBM 3081 mainframe.
- Compared to the mainframe, the workstation had to provide superior graphics and faster turnaround time, specifically in reservoir simulation.
- Software that needed to be run included black oil, steam and combustion, compositional, and CO2 flooding simulators.
- The workstation had to provide connectivity to Oxy's shared mainframe computer's data bases, as well as to the five local sites in the newly decentralized organization.
- Within each site, the workstation also had to allow networking with other workstations in a Netware 386 local area network (LAN), as well as with printers and plotters to provide reservoir simulation hard copy.
- Finally, each workstation had to be priced under $35,000.
Before the 1989 reorganization, when the entire simulation staff still was concentrated in Tulsa, MIS was prepared to recommend an Apollo DN10000. The proposed solution at that time was to use the Apollo as a compute server and to provide simulation engineers with desktop graphics workstations.
Following decentralization, the cost of the Apollo server/workstation approach was no longer competitive. Starting over, formal requests for vendor bids were sent out for evaluating a number of engineering workstation options. The IBM RISC System/6000 workstation, introduced to the market in 1990 soon after Oxy's reorganization, proved to be ideally suited for Oxy's simulation needs.
IMPRESSIVE BENCHMARKS
Through extensive benchmarking with Oxy's own applications, the RISC System/6000 Model 320 was determined to have the compute-power equivalent of the IBM ES/3081 and the IBM ES/3090-500E. In running various in situ combustion models, for example, the RISC System/6000 consistently performed at a level between the 3081 and the more powerful 3090.
Additionally, when the RISC System/6000 was benchmarked against the 3090 using an existing CO2 multiflood model, the workstation outperformed the larger mainframe. The simulation was produced in 225 vs. 294 min required on the 3090.
Software migration also proved to be no problem. Oxy was easily able to convert most of the existing applications. The majority, written in Fortran 77, ran virtually unchanged under AIX on the RISC System/6000 workstation.
These applications included:
- The computer modeling group's steam and combustion, and compositional simulation packages
- Todd Chase & Associates' CO2 flooding simulation system
- ECL-Berguson migration of its Grid preprocessor, Graph graphical postprocessor, and the Eclipse black-oil simulator to the RISC System/6000.
From a simulator run, the screen in Fig. 1 shows a predicted water saturation.
Oxy now has eight RISC System/6000 Model 320s and one Model 530, all networked with LANS. One of the 320s is in the Tulsa headquarters for software support, three 320s are in regional sites in Lindsay, Okla., Houston, and Oklahoma City, and four 320s and one 530 are in the Midland, Tex., regional office.
Since Oxy had no previous experience with UNIX workstations, IBM provided technical support and training.
MAINFRAME/WORKSTATION LINK
While all reservoir simulations are conducted on the workstations, the headquarters' mainframe continues to serve as a repository of historical reservoir data. These data are then transferred and downloaded, as needed, into the workstations for local processing.
In a typical simulation, geological information and the production history of oil wells residing on the mainframe are downloaded to the workstation to perform an historical match with the reservoir model being developed.
The mainframe data contain the monthly production results from similar wells and includes data regarding variables such as pressures, as well as oil, gas, and water volumes.
By using these mainframe data in the workstation simulation model, the engineer can refine the model and perform what-if analyses. These analyses enable the engineer to estimate the effects of working the field in different ways such as by drilling more wells, or employing CO2 or waterflood, and the volumes needed to effectively remove the oil.
The historical data dynamically refine the simulation models. When the simulation converges with historical reality, management uses the resulting information to determine whether to drill more wells or to use some form of recovery and to estimate costs.
GRAPHICS
Of course, computer graphics play a large role in reservoir modeling and analysis, simplifying the visualization of complex simulations. The high-resolution graphics on the RISC System/6000 literally give engineers a better picture for analysis. Currently under evaluation are new visualization packages to further enhance the capabilities of these systems.
Fig. 2 illustrates the pressure saturation in cross-section from a simulation run.
The workstation also provides significant improvement over mainframe response time, both in keyboard response and processing speed.
Faster keyboard response time allows graphics to be manipulated much more quickly after being generated. The workstation's greater speed in generating graphics simulations is particularly useful when the engineer is attempting to visualize a coherent process in action.
For instance, when the engineer wants to view the change of water saturation in a water injection process, workstation graphics allow the engineer to watch as the water migrates from the injection wells into the reservoir.
PAY BACK
The investment in workstations has paid off not just in improved graphics, but in overall processing costs. Previously, simulation managers would get a big surprise from MIS in mainframe charges each month. Now, their computing costs are dramatically reduced and fixed.
Initial expenditure for hardware and software was approximately $300,000. Subsequently, the only cost of the workstation-based simulation system is an annual maintenance fee of $17,000.
During the first half of 1990, one engineer racked up computing charges of $64,000/month on a CO2 miscible flood simulation involving heavy mainframe processing to achieve historical matches,
This processing was done at night at a discounted processing rate.
In saving the mainframe computing costs incurred by this one engineer alone, the workstation solution paid for itself in the first year.
Copyright 1991 Oil & Gas Journal. All Rights Reserved.
