By Maddy Reddy
Intel’s promotion of Itanium and Xeon as an alternative to RISC processors has been unrelenting and AMD has not been far behind with its Opteron offering. Initially, industries using RISC-based servers avoided the calls of the chip manufacturers and their partners. Now, however, commodity hardware running Linux appears to have broken down the doors.
|~|johnfoster1.jpg|~|“We are incorporating the InfiniBand and Myrinet technology into our server line for oil & gas customers,” says Sun's John Foster.|~|Intel’s promotion of Itanium and Xeon as an alternative to RISC processors has been unrelenting and AMD has not been far behind with its Opteron offering. Initially, those industries that required the power and scalability associated with RISC-based servers avoided the calls of the chip manufacturers and their partners. Now, however, commodity hardware running Linux appears to have broken down the doors.
Recovering oil from beneath the surface of the earth is a difficult job. Among other things, it takes months of geological study, large capital investment and advanced drilling techniques. Without such precision, millions of dollars can be wasted, as worthless dry wells are dug and reservoirs poorly exploited. Key to ensuring the process of extraction runs smoothly and doesn’t cost the earth is the processing of seismic data. Gathered by surveying the land with sound waves, the information is processed in high performance computing (HPC) environments that then deliver the data to visualisation centres, which scientists use to plan and guide drilling operations.
In the past, HPC environments have been run on high end proprietary servers, typically based on RISC chipsets and running a Unix operating system. Although effective, such machines are expensive to both buy and operate. Furthermore, as oil gets harder to find and more data has to be gathered and processed, the cost can only grow.
Such a dilemma forced Saudi Aramco, which produces approximately 25% of the world’s oil, to search for an alternative to its RISC-based, 64-bit servers to run its Kirchoff Prestack Time Migration (PSTM) — a set of compute-intensive algorithms that takes seismic data, enhances it, executes up to five million instructions per data point and assembles it into a usable image of an oil or gas deposit. “This is a mission critical application,” says Mohammad Huwaidian exploration systems analyst at Saudi Aramco.
“It does some of the most important work in our business. Some of the old ways we ran our PSTM just didn’t work well or didn’t have enough power. So we made a change,” he explains. This change manifested itself as a switch from proprietary RISC-based systems to commodity hardware running Linux. The first such cluster built by Aramco featured 900 IBM rack-mounted 1U nodes running Red Hat Linux 7.2 as its operating system. Each node had two Intel Pentium III processors running at 1.4 GHz and 2 Gbytes of memory.
Because the PSTM application is compute-intensive rather than communications-intensive, Saudi Aramco employs 100Mbit/s Fast Ethernet across its clusters. Saudi Aramco uses two messaging protocols to pass data around the cluster. Within each of the 900 nodes, the company uses OpenMP, an industry-standard API that supports multi-platform, shared-memory parallel programming on all architectures. For messaging among nodes, Aramco uses the Message Passing Interface (MPI) standard. “It’s extremely fast and a de-facto standard,” says Huwaidi.
By moving its PSTM application from a RISC-based platform and employing software engineers from Intel to optimise its code using the Intel Fortran Compiler for Linux 6.0 and Intel Vtune Performance Analyser, Aramco achieved a 33% performance boost while reducing the cost of its HPC environment.
The oil giant was also able to make decisions faster, reduce production costs and provide its geophysicists with a more accurate picture of where the oil lies. In turn, this increased the company’s success rate for drilling and reduced the time, money and effort wasted on false starts.
“In the end, we built the fastest yet least expensive PSTM system,” says Huwaidi. “Our ability to rapidly and inexpensively add computing power to our cluster was a major benefit,” he adds.
Following this initial installation, the cluster grew to 1800 CPUs by the middle of 2002 and was up to 2000 nodes by January last year. Since then, the Saudi oil giant has implemented clusters from other server vendors, including Dell, HP and Sun Microsystems.
Saudi Aramco is not the only oil & gas operator in the Middle East to invest in commodity hardware running Linux. AMD, which is targeting the oil & gas sector with its Opteron chips, has been on site with a number of the region’s energy giants, all of which it claims are showing an interest. Elsewhere, Intel has already been involved in a number projects. For example, Petroleum Development Oman (PDO) runs a Xeon-based IBM cluster and Abu Dhabi Marine Operating Company (ADMA-OPCO) operates an Itanium-based cluster of 24 Integrity servers from HP.
“There are a lot of Intel-based products going into the Middle East’s oil & gas industry in the HPC area. It is a RISC replacement and it started with Aramco a few years back and since then it has grown,” says Ferhad Patel, strategic relations manager for Middle East & North Africa at Intel. “If you look at most of the clusters going into the oil sector they are Intel-based. It seems to be the de-facto standard. It is now down to buying cycles. Most of the tenders issued specify commodity hardware,” he adds.||**|||~|ferhadpatel1.jpg|~|Most of the tenders issued today specify commodity hardware, says Intel’s Ferhad Patel.|~|The reasons for the growing adoption of commodity clusters in the oil & gas industry are very simple — it allows users to save money on their initial hardware investment, as well as reduce maintenance costs and improve performance levels. “Commodity clusters give a better price performance platform and enhance productivity for oil & gas companies, while saving them millions of dollars,” confirms Yasser Ragaei, sales manager for business critical systems at HP Middle East.“Intel-based systems give users speed and they allow them to get the results they are looking for faster. They also reduce maintenance,” adds Patel.
Moving forward, the key commodity server vendors are promising that these benefits will only improve as the likes of HP, IBM and Sun invest in clustering technology. For instance, both Sun and HP are working to improve the manageability of clusters. HP has already started shipping its Cluster Management Utility software, which allows clusters to be managed from a single point.
“Sun is also developing a number of cluster management solutions, including a couple of control station solutions developed at the request of clients in the oil & gas industry,” adds John Foster, volume solutions sales manager for Sun Microsystems in the MENA region.
Another key area for investment is improving the connectivity within clusters so they work even faster. Currently, these efforts are focused on Myrinet and InfiniBand. The former is a packet-communication and switching technology that achieves high performance by distributing demanding computations across an array of cost-effective hosts. InfiniBand, on the other hand, is the result of merging two competing designs — Future I/O and Next Generation I/O. It is an architecture and specification for data flow between processors and I/O devices that offers throughput of up to 2.5Gbytes/s and support for up to 64,000 addressable devices.
Both technologies are expected to replace Ethernet and eventually Gigabit Ethernet connections, thereby improving the performance of clusters. Server vendors are already looking to accommodate these standards into their hardware offerings.
“Sun is incorporating InfiniBand technology in its server platforms, application environments, switches and storage for customers in the oil & gas industry, which helps to reduce processing time and increase scalability. Sun is also integrating Myrinet technology [in its servers] to support clustering initiatives, which provides faster processing within clusters,” says Foster.
AMD is hoping to use the connectivity issue to overhaul its rival Intel. According to Pierre Brunswick, AMD’s regional sales director for Russia-CIS, Middle East & Africa, the Northbridge and HyperTransport technology found in the vendor’s Opteron chip sets it apart because it facilitates point-to-point connectivity and ensures there is no latency to the memory. Furthermore, Opteron’s HyperTransport already connects to both Myrinet and InfiniBand, thus accelerating processing in HPC environments.
“HyperTransport is in the system and it has a very fast interconnect with protocols like InfiniBand and Myrinet which speeds up clusters and makes them faster,” explains Brunswick.Regardless of the advances being made in clustering technology and that commodity hardware looks set to dominate the oil & gas industry’s HPC environments, it has yet to penetrate other areas of the sector. In particular, the visualisation environments, which allow energy companies to render, manipulate and work with the seismic data processed in the HPC farms, remain proprietary.
Khalda Petroleum Company (KPC), for example, runs its Landmark applications on Unix. According to the company’s IT manager, Amira Greis, it is unlikely to consider commodity clusters in this environment for some time. “We wouldn’t look at something like Linux on commodity hardware because oil & gas applications for this environment are still quite new,” she says. “When you buy software for millions of dollars you do not look for a cheap hardware platform,” she adds.
Intel’s Patel also blames the lack of commodity hardware adoption in the visualisation arena on the costs associated with such set ups. “Oil companies do not mind buying some proprietary systems because of the return on investment (ROI). For example, visualisation centres need proprietary systems and although they cost between US$1million to US$3 million they can save a company having to drill a US$20 million dry well that doesn’t deliver,” he says.
Such reasoning has so far prevented local adoption of Intel or AMD based clusters running Linux in visualisation centres. However, this could be about to change as Schlumberger Information Solutions (SIS) has developed a visualisation application capable of running on commodity clusters. Called GigaViz, the solution was acquired when SIS bought VoxelVision. Although today it is only commercially available as a standalone application running on Microsoft Windows at the client and Linux at the backend, this will change next month as SIS unveils a new version of the software.
The improved GigaViz will be built directly onto SIS’s GeoFrame. As a result of this integration, the product will support Linux on the client side as well as on the cluster. According to Russ Sagert, reservoir software products business development manager at SIS, this will provide users with a number of benefits, including enhanced performance and reduced costs.
“The commodity platforms are cheap enough to allow the user to buy more [processing power] for the same amount of money. Combine this with each processor and graphics card being faster as well as cheaper, then the user’s budget now buys substantially more computing and graphics power than ever before,” he says. “With GigaViz we can combine the performance and scaleability of these Linux clusters with their better price points,” Sagert adds.
Furthermore, it appears that the Middle East’s oil & gas players are already taking an interest in GigaViz. The Petroleum Institute of Technology in Abu Dhabi has already purchased it, exposed the entire ADNOC group of companies to it, and the likes of ADCO, ADMA and ZADCO are supposedly preparing to evaluate it on their own.
“During the Geo2004 industry trade show in Bahrain, GigaViz drew huge crowds of people wanting to see this new technology. Companies that do their own seismic processing, like Saudi Aramco, showed keen interest,” says Sagert. “[Linux and clusters] are a trend that has become prevalent in other industries like the manufacturing, medical, and entertainment industries, so we expect it will start doing the same in the oil & gas industry. With the benefits in price, performance, and saleability, it is just a matter of time,” he adds.
However, while SIS is obviously keen to see GigaViz succeed, other are more cautious with regard to pinpointing when exactly the oil & gas industry will embrace commodity clusters in the visualisation environment. This is due to the applications’ lack of maturity compared to those that have been running on proprietary platforms for many years.As such, Gilbert Soufan, territory manager for SGI in the Middle East, believes proprietary systems will continue to remain the de-facto visualisation platforms in the short term. In the medium term, he believes a hybrid model, such as SGI’s Linux-based Altix machine, could take hold.
“We believe that the oil companies are very interested in an approach that is not proprietary but it is not the cheap commodity products either. It gives them the best of both worlds because it is a hybrid that is specific for their heavy computational needs,” Soufan says. ||**||