The ability to sift through mountains of data is critical for businesses in the petroleum industry. At Petroleum Geo-Services (PGS), an Oslo, Norway, firm with 2,000 employees and revenue of approximately U.S. $1.5 billion, the data streams in 24×7 from a dozen offshore vessels and 30 offices worldwide, which include locations in Europe, Asia and the Americas.
PGS, which operates nearly two dozen data processing centers, helps other firms find offshore oil and gas reserves all over the world. At the center of the company’s business model are seismic and electromagnetic services, data acquisition, data processing, reservoir analysis and interpretation, and multi-client library data.
“The oil and gas industry has always been at the leading edge of high-performance computing,” says Guillaume Cambois, executive vice president, Imaging & Engineering, at PGS. “It is the largest non-government user of HPC. Seismic surveys are ever increasing in size and data density, and imaging algorithms are becoming more accurate and more complex.”
Consequently, PGS turned to a Cray XC40 supercomputer to improve processing for a variety of complex tasks. These include generating clear three-dimensional images of the subsurface geologic structures from sound recordings that are later used to guide drilling.
The imaging algorithms manipulate and properly position the peaks and troughs of billions of recorded sound waves to create a picture of subsurface geologic structure. The image is later used to identify likely oil traps and guide drilling.
“The accuracy of the algorithms and the resolution of the image depend on how much compute power we can throw at the problem,” Cambois points out.
PGS, which has previously relied on HPC’s from a variety of other vendors, opted for the Cray XC40 because management believed that the vendor “has made significant advances in the network fabric with its Aries network,” he notes. The company switched on the new system in May.
Improving Performance and Scalability
The network topology frees applications from locality constraints and provides a number of other features that improve performance and scalability. Cambois says that the company is migrating away from mainframes and toward Beowulf clusters.
“With the increase in processor density in today’s clusters, the throughput bottlenecks have moved from compute power to data movement,” he explains. “The Aries network creates a tightly integrated node cluster that allows for a distributed memory approach to our imaging algorithms.
“The ability to spread our image over the memory of multiple nodes—while still providing fast access to the memory from any node—will allow us to make a leap forward in imaging technology.”
Cambois reports that PGS is now looking to chip makers to boost memory bandwidth, to switch makers to support even faster networks and to Intel to further integrate network fabric at the silicon level.
“With the recent introduction of PGS Titan-Class Ramform vessels, seismic surveys are larger and denser than they have ever been,” Cambois points out. “This trend will continue, along with our clients’ demands for clearer images of the subsurface, and therefore more accurate and complex algorithms. Ever-increasing data volumes will continue to drive seismic data processing in the future.”