SUPERCOMPUTING IS

IBM’s BUSINESS
 

by Nancy Cohen

     

 

“From manufacturing to financial services to petroleum, virtually every company can benefit from some form of high-performance computing.” 

– David Gelardi, director, High Performance Computing, joined IBM management in 1984 with subsequent focuses on Unix, AIX, and RS/6000. He leads in driving sales of high-performance computing and the pSeries HPC line, which includes Linux and AIX-based servers.
In the past, looking for voices at IBM to address clustered implementations for supercomputing meant flitting around a number of units and product groups. Maybe you found your questions answered after cornering a server-benchmarking higher-up. Maybe you sought senior technologists at Deep Computing. Maybe you hung around supercomputing gurus at IBM Research. 

The fact is that IBM, ever morphing units to whatever market realities govern, has raised its positioning of high- performance computing as a one-stop oracle for replies: Rather than delegating high-performance computing to be a subset of some discrete research or product or industry unit, it’s now a top umbrella for IBM’s targeting to transform server, software, and Linux technology showcases into sales.

David Gelardi, director of high-performance computing at IBM, is driving IBM’s new look in supercomputing.  “We recently did some reorganization. We looked at the high-performance computing community and customers
through their vantage points, and we chose to use high-performance computing as a superset for all our computationally intensive areas in the organization.” Translation: IBM is out to capture a palpable market opportunity in the name of high-performance computing. If supercomputing clusters were divided into two user camps: Research/academia and the tomorrow-land of enterprise use of computation-intensive clustering, then “tomorrow” is here. 

High-performance computing servers have come to public attention largely from scientists and university researchers using it for modeling and predicting how the planets turn, or how crops can grow, or which diseases can be eradicated, or what cells can reveal. Today, the oil industry is only one of a number of industry segments that IBM sees as ripe for use of clustered supercomputing to achieve complex calculations on massive amounts of data. Stepping beyond the steep investment specter in supercomputers, users are turning to clustering, or grouping large numbers of low-cost servers, to come up with single powerful computing systems.

Consider the spate of IBM announcements all this year [see box]. A high-performance push is bringing in results in university/R&D/scientific markets while driving home a message that supercomputing for computation-intensive reasons not just high availability is what business organizations today can use for competitive advantage.

“We know that talking about products not just in and of themselves as big iron but in the light of the key segments in which they are placed, turns corners well,” says Gelardi. “From insurance, to manufacturing, to financial services, to petroleum, virtually every company today has transactional needs, and can benefit from some form of high-performance computing. It might be a small part of the enterprise or a great part of the enterprise. I can look at insurance companies to find them doing number of computational things. I can go to Wall Street and see risk arbitrage at work. Businesses are running complex models and they’re making complex decisions.”

Gelardi notes that one easily thinks of database implementations as the key concern of large organizations. “The database is not the only issue. Auto companies have huge computational needs. In car manufacturing alone, crash analysis is very compute-intensive, where they need to run automobile deconstruction.”

Dipping into IBM’s recent past, Gelardi holds the Boeing design story as a real harbinger of how supercomputing is catching on in business. (Boeing has been aggressively seeking leadership branding in applying advanced technologies to manufacturing operations. IBM partnered up in a number of their projects. Together with Dassault Systemes, Boeing demonstrated the value of a digital mock-up, initiated with the 777 and then adopted by other Boeing programs, when they began talking in terms of “virtual product development.”

Of numerous releases this year on supercomputing and Linux clusters, a most eye-catching IBM announcement is the Intel Itanium story: IBM’s new Itanium-based eServer, part of the Linux cluster at the NCSA, is capable of 1 trillion calculations per second, and is to be used for scientists studying a range of questions, from how gas flows to black-hole collisions. 

Itanium is Intel's next-generation, 64-bit processor family as well as for the first chip in that new generation. Computer manufacturers are to introduce initial Itanium-based servers and workstations. Intel says up to 25 vendors are expected to start shipping servers and workstations based on the 64-bit processor. They include Dell Computer, IBM, Hewlett-Packard, and Compaq Computer. 

The Itanium processor’s Explicitly Parallel Instruction Computing (EPIC) design enables breakthrough capabilities in processing terabytes of data, speeding protected online  purchases and transactions, and processing complex computations. Clearly, applications for business environments. What does that mean for IBM? A sales message with Itanium positioned as the magic bullet for e-business? 

No. Gelardi says that IBM is tailoring Linux clustering to whatever hardware and software distributions work for the individual customer. “The difference in Itanium is that it is a stronger machine when you want to run floating-point base applications. Not every corporation needs to do that. Itanium is more expensive than IA-32 and there is no sense to spend a lot of money if you do not want it.”

The Itanium processor’s Explicitly Parallel Instruction Computing (EPIC) design enables breakthrough capabilities in processing terabytes of data, speeding protected online  purchases and transactions, and processing complex computations. Clearly, applications for business environments. What does that mean for IBM? A sales message with Itanium positioned as the magic bullet for e-business? 

No. Gelardi says that IBM is tailoring Linux clustering to whatever hardware and software distributions work for the individual customer. “The difference in Itanium is that it is a stronger machine when you want to run floating-point base applications. Not every corporation needs to do that. Itanium is more expensive than IA-32 and there is no sense to spend a lot of money if you do not want it.

”WesternGeco, for example, which chose an IBM Linux supercomputer to power sophisticated seismic imaging, is running a Linux cluster on existing Intel IA-32.  And at Max Planck, the eServer involved is a cluster of Power4 miniprocessors running AIX. Gelardi says that some scientific company investments have been heavily oriented toward Linux and Intel IA-32 and not -64, in cases where applications are more integer-intensive than floating-point intensive.

In instances where Itanium presents a strong alternative, the scenario will be watch-and-wait. “Banks will not rush out to invest in Itanium very early on. Banking’s IT planners will wait for universities, research, and scientific arenas first and watch results. There’s an 18-month trickle-down generally, when you trace the move of a technology from experimental to commercial.

  

  IBM’s Year of Supercomputing

January 2001

The NCSA says it will install two Linux clusters with IBM eServers running Linux and Myricom cluster interconnect network. NCSA and IBM announce their partnership to create world’s fastest Linux supercomputer in academia.

  



  May 17, 2001

IBM announces Germany’s Max Planck Society for Advancement of Science gives IBM green light to build Europe’s largest non-classified supercomputer.

  



 

May 25, 2001  

IBM announces WesternGeco win, in the latter’s move to IBM Linux supercomputing (xSeries-based Linux clusters) in survey work for global oil companies.

  

 
 
  May 29, 2001

IBM announces Itanium-based eServer and Intellistation workstations to form supercluster at NCSA at University of Illinois Urbana-Champaign. 

  
 
           
” The real door-breaking story in supercomputing clustering moving to business is in fact the oil industry. Gelardi’s chapter one begins with Royal Dutch Shell when the latter decided to go with a clustering implementation late last year. The oil exploration unit of Royal Dutch Shell Group Inc. said it’s working with IBM to build a Linux-based supercomputer linking 1,024 servers Shell was to use the clustered system to analyze seismic data and other geophysical information as part of its efforts to find new supplies of oil. 

Oil producers need considerable processing power to analyze the massive amounts of data collected in underground searches, in which explosions are set off in an attempt to determine the location of oil reserves based on sound waves collected by sensors. “This was one of our largest clusters,” he says. “They are running a gigantic job farm, where they are gathering data for potential sites for oil. They run sophisticated algorithms that analyze collected land data, for better decisions that the thing you think might be oil is actually not water or natural gas. They use very large clusters.”

Beyond the oil industry, says Gelardi, “The year 2001 is an important one for Linux clustering becoming mainstream. If you look at the industry over time, there have been lots of interesting technologies in high-performance communities that never went beyond those communities. “I am excited about Shell and WesternGeco saying they can bet their business on Linux clustering.”
  BEOWOLF ALLIANCE  


 

Compaq, like IBM, is selling its Linux clustering expertise in building high-performance technical computing clusters. Compaq announced last month that it was partnering with strong Linux Beowulf clustering players to deliver Beowulf clustering solutions on its ProLiant server platform. Market targets: customers from oil and gas, aerospace, finance, biotechnology sectors, and just about any small, medium, or large enterprise that wants to achieve supercomputing power economically via building clusters that range from 16 to over 512 nodes. 

Companies listed in the Compaq alliance announcement include San Diego Supercomputer Center's NPACI Rocks; Scientific Computing Associates’ Linda and Paradise middleware; Scyld Computing Corporation’s Beowulf Operating System; Turbogenomics’ Turbo Blast, and TurboHub for bioinformatics analysis. 

  
           
  Case in Point:
WesternGeco’s clustering 		    

Will the clustering approach be the one and only sensible road to harness crunch-intensive computing? Not every business planner feels as confident. Some analysts believe that there is no substitute for vector supercomputers for certain  applications and that vector computing remains as a viable alternative for compute-intensive tasks. The Cray SV2 series, for example, is described as an extreme-performance vector supercomputer, and it is intended for availability in the second half of next year. 

“Oh, there’s no question there are classes of applications better served by vector capabilities,” says Gelardi. “But vector computing is narrowly available, while clustering has become prevalent. And as a customer, I am able to see that clustering gives me tremendous freedom and flexibility where I am no longer tied to the vector instruction set.”

Gelardi says IBM can say ‘been there done that’ with vector computing: “We moved to a message-passing world and then we carried that forward to our Linux clustering effort.” Observers at Linux NetworX, meanwhile, say clustering concerns in the enterprise have less to do with the technology per se than with concerns about who and how to manage the cluster that’s to be adopted. Having the right management tools will become the real door-opener for broader adoption.

Sure, Linux clustering technology has proven to be a powerful and cost-effective alternative to traditional supercomputers, says CEO Glen Lowry, but the question for the organization planners remains: How manageable is the cluster? “With the right cluster management tools, organizations can dedicate their resources to number crunching or serving instead of cluster administration,” says Lowry. “Linux clustering is such a powerful and reliable technology,” that with the right management tools “enterprise businesses have every incentive to adopt the technology.”

As for Linux distributions, Gelardi says “We’re seeing a lot of Red Hat followed by TurboLinux.” Still, Gelardi hastens to add that IBM is distribution-neutral. “Our role is to promote open access of Linux. We won’t impose anything other than what is best for customers.” 

 

In the global world of oil exploration, high-performance clustering at an economical price is a major gain for WesternGeco, in its decision to use a Linux cluster of 256 nodes. WesternGeco is in the business of seismic imaging. Oil companies are using seismic imaging provided by companies like WesternGeco to save big bucks—very big bucks. 

When an unsuccessful drilling effort costs millions, the obvious option is to choose a service provider who can do seismic imaging instead, where scientists can provide maps of oil and gas reservoirs so that explorers can better see where to drill before the drilling even starts. WesternGeco, an aggressive worldwide player competing in the field of imaging services, with corporate office in Gatwick and large facility in Houston, came into being earlier this year after a morphing of Geco-Prakla and Western Geophysical companies. Performing these multi-client seismic surveys requires sophisticated computation analysis—the stuff of which supercomputers can do, but at an expense that competitive companies in this field seek to avoid. 

Trevor Gatus, WesternGeco’s Houston land manager, who manages the Houston data processing facility, says “We always had the capacity to do imaging from a mathematical perspective, but we were restricted: We looked for more cost-effective approach than with expensive supercomputing machines.” In May, IBM announced it had won the deal: WesternGeco went for a cluster of 256 IBM eServer xSeries systems running Linux to do intense computations for seismic imaging. The systems in the cluster are powered by two Intel Pentium III processors.

“We are now in the process where we are to bring it on-line in several months,” says Gatus. The payoff in being able to do supercomputing at a rapid and economical clip will be obvious in an industry where virtually all drilling expeditions are now said to be preceded by seismic imaging. How unique is WesternGeco in its clustering implementation? “There are other geophysical contractors using clusters,” says Gatus. “Seismic acquisition is an expensive ballgame to play in and we are all on the lookout for an edge.” 

The victory for WesternGeco: Turnaround. When you think of what was the turnaround norm 20 years ago, the push forward enabled by clustering is impressive. Gatus says 20 years ago, jobs took “months and months.” Eventually, the turnaround was about eight weeks. And now: “The use of IBM systems running Linux has greatly expanded our ability to evaluate drill sites,” he says. “With better price/performance, we can complete jobs that once took eight weeks in three.”

As for maintaining the cluster approach, is there any fear that the complexity of it all will be too much to handle? “We do this stuff for a living,” answers Gatus. “We have been haggling with computer systems every day since we first did data processing.”