Print this article

Comment

01 September 2003

Linux elbows into super-computer turf

Amerada Hess, WesternGeco change seismic oil exploration systems.

By Jim Strothman

Linux-based systems have gained a foothold in a traditional supercomputer stronghold: seismic oil exploration.

Oil and gas giant Amerada Hess's Houston-based Exploration department switched to Linux for three-dimensional (3-D) modeling used to find oil and gas in hard-to-reach underwater areas. Separately, WesternGeco, jointly owned by Schlumberger and Baker Hughes, also turned to Linux to produce 3-D images of the earth's subsurface.

Frustrated with the high costs and difficult upgrades of traditional supercomputing hardware used to process terabyte-sized exploration data volumes, Amerada Hess ported its IBM supercomputing software to Linux-based software servers. The company replaced its $2.5 million IBM system with a $420,000 Red Hat cluster.

Jeff Davis, Amerada Hess senior systems programmer, said the Linux system proved easier to upgrade and install. It is also several times more productive, he said.

"We started with IBM 3090 mainframes," Davis said, "Then, about 1993, we decided the IBM mainframes were too expensive and we would go with a distributed processing system. We used an IBM SP [scalable processor], one of the Deep Blue [supercomputer] machines that played chess. However, it was costing us about several hundred thousand a month for lease and maintenance."

About 1998, Hess decided to look at Linux, the senior systems programmer said. It started a pilot project with Paralogic, a supplier of turnkey, Linux-based parallel computing systems. Davis said Hess successfully ported its code to Paralogic in a matter of a few weeks.

"For about the cost of one month of the [IBM] SP system, we could buy a Paralogic system," he said. Hess brought in a Paralogic system and successfully put it into production. However, Davis and others decided "we could do it cheaper if we did it ourselves," he said.

POWERED BY PCS

Hess initially bought 32 Dell Precision (dual-processor) Workstation 410s to power its 32-node Linux Beowolf cluster. The workstations proved compatible with Amerada Hess's initiatives, and the company ordered 32 additional units, plus a PowerEdge 1300 server to act as a file server for its Linux environment.

The exploration department installed Red Hat's kick-start software and found the Dell/Red Hat combination could do all that the Paralogic system did—at about 25% less cost, Davis said. The Paralogic system cost Hess $160,000. "We did it with Dell for about $115,000."

Davis said the oil company now has about 248 dual Dell processors, or the equivalent of 496 CPUs. Throughout the process, Hess didn't have to add more staff programmers, he said.

The cluster of Linux workstations performs mathematical functions for Hess's mission-critical exploration activities.

"The biggest problem with supercomputers is they're extremely expensive," said Davis. "Using Red Hat's software, we were able to take commodity hardware and easily build a supercomputer ourselves."

Amerada Hess also uses the Red Hat operating system on preconfigured Dell servers for two-dimensional geophysical modeling, IBM 3590 tape access, and Web and database applications.

Almost all of Hess's exploration data processing software was written in-house, Davis said, where about five programmers maintain all the applications. By using off-the-shelf PCs, "we're no longer dominated by hardware vendors."

The department handles "millions and millions of lines of code," he said, generally processing data from the floor of the Gulf of Mexico. Offshore contractors typically have a battery of microphones that collect noise data fed to high-capacity IBM tape drives. "We're talking terabyte-sized volumes. One job may run a week, twenty-four hours a day," he said.

"We get the fastest [machines] Dell has at the time. All are dual CPU systems: 1 gigabit of memory, 2-gigabit hard drives on a switched network." The systems "crunch numbers in parallel," reading data off the IBM tapes and exchanging results with one another in parallel. A master computer breaks the 3-D cube into pieces, and each worker computer handles a piece of the data.

WESTERNGECO TURNS TO LINUX

WesternGeco, a global provider of integrated exploration and reservoir imaging services to oil companies worldwide, also recently installed a Linux-based system. A joint venture between Schlumberger and Baker Hughes, WesternGeco, headquartered near London, produces 3-D images of the earth's subsurface. Based on data collected in seismic surveys, the 3-D images enable drillers to accurately locate hydrocarbon reserves and assess their size.

To compete with other seismic services companies, WesternGeco needed massive computer processing power to produce images quickly and at the lowest cost, said Kannan Venkataraman, WesternGeco area manager, worldwide computer systems and support.

The exploration services firm was attracted to Linux's scalability and reliability, as well as to the parallel processing speed of clustered servers. It decided to supplement its IBM RS/6000 SP and SGI technology with a Linux cluster of Intel-based servers.

Housed at the Houston facility where WesternGeco's data processing division is located, the cluster consists of 256 IBM eServer xSeries servers with dual Intel 933-megahertz Pentium III processors running Red Hat Linux.

Installation then began on a separate cluster at the company's Bedford, England, location with 512 xSeries nodes. WesternGeco uses 100Base-T Ethernet switching to interconnect the nodes and a gigabit Ethernet switch to link the "master" server to the data source.

PCPC Inc., an IBM business partner specializing in Linux systems, racked and delivered the servers and worked with WesternGeco's information technology staff to install Linux and the networking hardware. IBM provided a three-year hardware and software support warranty.

Eventually, Venkataraman said, Western-Geco may add other IBM storage and networking equipment to supply the cluster with high-capacity storage capabilities. "By adapting its products to Linux, IBM is essentially saying that Linux is ready for enterprise applications. And we've proven that here at WesternGeco," the computer systems area manager said.

Venkataraman said Linux, in combination with the servers, has proven robust enough to handle the computationally intensive seismic processing applications. The Linux-based system resulted in not only faster processing and better customer service but also lower costs, he said.

"The IBM Linux cluster, built with more commodity parts, is far more price competitive than any alternative solution providing equivalent processing capacity," Venkataraman said. IT