Supercomputers To Enable Safter, More Efficient Oil Drilling

For decades, theindustry has used computers to maximize profit and minimizeenvironmental impact, explained Tahsin Kurc, assistant professor ofbiomedical informatics at Ohio State University.

Typically,companies take seismic measurements of an oil reservoir and simulatedrilling scenarios on a local computer. Now Kurc and his colleagues aredeveloping a software system and related techniques to letsupercomputers at different locations share the workload. The systemruns simulations faster and in much greater detail – and enablesanalysis of very large amounts of data.

The scientists areemploying the same tools and techniques that they use to connectcomputing resources in biomedical research. Whether they are workingwith images from digitized microscopes or MRI machines, their focus ison creating software systems that pull important information from theavailable data.

From that perspective, a seismic map of anoilfield isn't that different than a brain scan, Kurc said. Bothinvolve complex analyses of large amounts of data.

In anoilfield, rock, water, oil and gas mingle in fluid pools undergroundthat are hard to discern from the surface, and seismic measurementsdon't tell the whole story.

Yet oil companies must couple thosemeasurements to a computer model of how they can utilize the reservoir,so that they can accurately predict its output for years to come. Andthey can't even be certain that they're using exactly the right modelfor a field's particular geology.

“You never know the exactproperties of the reservoir, so you have to make some guesses,” Kurcsaid. “You have a lot of choices of what to do, so you want to run alot of simulations.”

The same problems arise when a company wantsto minimize its effects on the environment around the reservoir, ortrack the path of an oil spill.

Each simulation can require hoursor even days on a PC, and generate tens of gigabytes (billions ofbytes) of data. Oil companies have to greatly simplify their computermodels to handle such large datasets.

Kurc and his Ohio Statecolleagues – Joel Saltz, professor and chair of the Department ofBiomedical Informatics, assistant professor Umit Catalyurek, researchprogrammer Benjamin Rutt and graduate student Xi Zhang – are enablingtechnologies to spread that data around supercomputers at differentinstitutions. In a recent issue of the journal Concurrency andComputation: Practice and Experience, they described a software programcalled DataCutter that portions out data analysis tasks among networkedcomputer systems.

This project is part of a larger collaborationwith researchers at the University of Texas at Austin, Oregon StateUniversity, University of Maryland, and Rutgers University. Theinstitutions joined together to utilize the TeraGrid network, whichlinks supercomputer centers around the country for large-scale studies.

Programslike DataCutter are called “middleware,” because they link differentsoftware components. The goal, Kurc said, is to design middleware thatworks with a wide range of applications.

“We try to come up withcommonalities between the applications in that class,” he said. “Dothey have a similar way of querying the data, for instance? Then wedevelop algorithms and tools that will support that commonality.”

DataCutter coordinates how data is processed on the network, and filters the data for the end user.

Theresearchers tested DataCutter with an oilfield simulation programdeveloped at the University of Texas at Austin. They ran threedifferent simulations over the TeraGrid: one to assess the economicvalue of an oilfield, one to locate sites of bypassed oil, and one toevaluate different production strategies – such as the placement ofpumps and outlets in an oil field.

The source data came fromsimulation-based oilfield studies at the University of Texas at Austin.That data and the output data from the simulations were spread aroundthree sites: the San Diego Supercomputer Center, the University ofMaryland, and Ohio State.

Using distributed computers, they wereable to reduce the execution time of one simulation from days to hours,and another from hours to several minutes. But Kurc feels that speedisn't the only benefit that oil companies would get from doing theirsimulations on computing infrastructures such as TeraGrid. They wouldalso have access to geological models and datasets at memberinstitutions, which could boost the accuracy of their simulations.

TheNational Science Foundation funded this project to make publiclyavailable, open-source software products for industry and academia, sopotential users can download the software through an open sourcelicense and use it in their projects.