IBM Think Research | The Blue Gene revolution
1 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
Think Research Home chip technology data management deep computing e-business technologies & services exploratory research hardware interfaces & accessibility network technology pervasive computing security servers & systems management
The Blue Gene revolution
services research software storage TR Print Archive Research Publications Research News
The world's most powerful supercomputer by Richard Silberman During the past five years, IBM Research has inv enormous resources toward developing a new fa supercomputers. Known as Blue Gene, these ma scalable machines are built with a radical new ar that makes them more powerful than anything se and promises to revolutionize high performance By making dramatic reductions in power consum and space requirements, IBM researchers and en helping turn ultra-powerful computing into an af practical and widely accessible tool for science, industry. The largest planned machine will be completed e It is funded by the U.S. Department of Energy's N Nuclear Security Administration and is being bu collaboration with the Lawrence Livermore Nati Laboratory. However, IBM is already assemblin that measures one quarter the size of that machin qualified as the world's most powerful supercom according to the Top500 list of supercomputers. IBM's Blue Gene/L machine achieved that top ra November-replacing the NEC Earth Simulator, w claimed the top spot three years ago. The Blue G
19/04/2005 11:56
IBM Think Research | The Blue Gene revolution
2 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
supercomputer has a sustained speed of 70.72 teraflops, which nearly doubles the Earth Simulator's previous top mark of 35.86 teraflops (one teraflop equals one trillion mathematical operations per second). This is an incredibly powerful computer and a major milestone for the Blue Gene project. However, to get a sense of the extraordinary computing power on the horizon, one must consider that the finished Blue Gene/L system will be four times larger than this prototype with a peak performance of 360 teraflops. "Blue Gene/L will likely have more processing power than today's 18 fastest supercomputers combined when complete," said Tilak Agerwala, vice president, systems, IBM Research. "With its innovative design and unprecedented scalability and flexibility, Blue Gene will have a profound impact on how supercomputers are built, who can use them, and what we can achieve with them." The Blue Gene project is part of IBM's ongoing commitment to push the limits of computing power and lead the world in supercomputer development. Seven years ago, IBM demonstrated the capabilities of high performance computing with Deep Blue, which defeated chess grand master Garry Kasparov. Subsequently, IBM built its ASCI White machine, which was 1,000 times more powerful than Deep Blue. While both these machines represented groundbreaking efforts for their times, Blue Gene/L stands in a league of its own. "Deep Blue had 32 processors and with Blue Gene/L we are talking about thousands of processors," said Gyan Bhanot, who is in charge of external application development for IBM Research. "This new machine is really more of a revolution in design than an evolution of IBM's previous supercomputer designs-and it clearly marks the beginning of a new era in supercomputing." A wide range of uses-from protein folding to financial modeling As powerful as conventional supercomputers have become, many scientific and data analysis problems require computational power much greater than what is presently available. The complex simulation and modeling required to solve many challenges in biology, medicine and other fields simply cannot be done on existing machines because of insufficient computing capabilities. Blue Gene is going to change all that. One of the most important areas where Blue Gene will prove invaluable is modeling the folding of human proteins-long considered one of science's Grand Challenges. Many deadly 19/04/2005 11:56
IBM Think Research | The Blue Gene revolution
3 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
diseases, including cystic fibrosis, sickle-cell anemia, Alzheimer's disease and "mad-cow" disease are thought to arise from misfolding proteins. Unlocking the mystery of this phenomenon will help researchers better understand diseases and could provide insights for those working on potential cures. "In biology, we are trying to understand processes, like protein folding, that occur at a submicroscopic level and therefore cannot be observed through a microscope," Agerwala said. "Blue Gene will help make these submicroscopic events visible in a way previously impossible by simulating them computationally." Blue Gene's massive computational power will open the door to an extraordinary period of progress and understanding not just in the life sciences, but in a range of other scientific and commercial fields, including material sciences, quantum chemistry, molecular dynamics, fluid dynamics, climate modeling, seismic exploration, aircraft and automotive design and financial risk analysis. Researchers at the Lawrence Livermore National Laboratory plan to use their 360 teraflop Blue Gene/L to investigate cosmology and the behavior of stellar binary pairs, laser-plasma interactions, and the behavior and aging of high explosives. ASTRON, a leading astronomy research group in the Netherlands, will use a 34-teraflop Blue Gene/L supercomputer to study the moments surrounding the creation of the universe (the Big Bang). Others, including the University of Edinburgh and the San Diego Supercomputer Center, have also signed contracts to acquire smaller IBM eServer Blue Gene/L systems of their own. "Blue Gene/L can scale up to a size large enough to tackle just about any problem in any field that requires massive crunch power," Bhanot said. Unprecedented time scales, faster research The Blue Gene project is not just about designing and building a new family of supercomputers; it also has a vigorous, innovative research program behind it. IBM is already performing scientific research on early Blue Gene/L prototypes. For instance, IBM researchers are currently studying G-protein coupled receptors (GPCRs), a large class of membrane-bound proteins that are important drug targets. Research teams are focused on simulating rhodopsin, a GPCR found in the retina, in hope of understanding how lipids reorganize in the presence of this protein. One obvious benefit of Blue Gene/L is how much faster it can perform these sorts of simulations than standard supercomputers. With conventional computing power, it can take weeks, months or more than a year to run detailed
19/04/2005 11:56
IBM Think Research | The Blue Gene revolution
4 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
simulations of submicroscopic processes, depending on how complex the process is and the duration of the process you're trying to simulate. Using Blue Gene, IBM researchers are already seeing performance improve by a factor of ten over existing commercial supercomputers-and this should reach a factor of 100 as larger Blue Gene systems become available. Nanosecond simulations that took four days to run on standard 48-node parallel computer systems currently require just 14 hours on a small BlueGene/L prototype. The full size Blue Gene systems, currently under construction, will enable researchers to routinely simulate microsecond time scales-one thousand times longer than currently possible. The centerpiece for IBM's continued application development for Blue Gene/L will be a 20-rack supercomputer that will be installed at the IBM Thomas J. Watson Research Center by March 2005. At a peak performance of more than 100 teraflops, this system is expected to be the largest privately owned supercomputer in the world. "This massive leap forward in speed will dramatically change the way scientists work and will have a huge impact on scientific discovery," said Bob Germain, manager of biomolecular dynamics and scalable modeling at the IBM Thomas J. Watson Research Center. With Blue Gene, scientists will be able to run simulations over much longer periods to understand long-term effects. High-performance computing will become a truly interactive tool; instead of waiting weeks for results, researchers will be able to run a calculation in an hour, see the results, make adjustments, and run it again. "We will be able to test scientific models on a scale that they have never been tested before, study phenomena that no one has ever studied before, and complete research in record time," Germain said. "Blue Gene will also reduce the need for costly and often dangerous physical experiments in many fields because now it is practical to do the necessary work computationally." Massive compute power, minimal floor space Blue Gene/L was designed to provide unprecedented levels of computing power, but it also costs less, consumes less power and takes up far less floor space than any comparable supercomputer-making this kind of computing power accessible to a much wider community of researchers and enterprises. Whereas many of today's supercomputers require their own buildings and power supplies, the 70.72 teraflops Blue Gene/L system occupies just 16 refrigerator-sized racks. The completed 64-rack Blue Gene/L machine will be four times faster, yet consume three times less power and ten times 19/04/2005 11:56
IBM Think Research | The Blue Gene revolution
5 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
less space than today's fastest supercomputers. "The challenge of computer design in the 21st century will be energy efficiency," said Eric Kronstadt, IBM Research's director of exploratory server systems. "Blue Gene's highly efficient design enables it to pack more compute power per cubic centimeter, by far, than any other computer in its class." "A university or business that needs massive computing power will be able to install a few racks of Blue Gene/L in an office-sized room and then scale up as desired," Agerwala added. "Blue Gene/L scales up quickly and provides a capability to solve hard problems, beyond traditional Linux clusters." The system uses thousands of powerful nodes and an innovative system-on-chip technology that enables it to scale to the highest tiers of performance while maintaining high reliability. An institution or enterprise that deploys a Blue Gene/L machine will be able to deal with increasing demand by scaling upwards to hundreds of thousands of processors. With conventional clusters, users often encounter scalability problems after a few hundred processors. Blue Gene/L's high degree of scalability is illustrated by the fact that in November 2003, a 512-node Blue Gene/L prototype was ranked the 73rd most powerful supercomputer in the world. One year later, a 16,384-node machine is now the world's most powerful supercomputer. The finished Blue Gene/L delivered to Lawrence Livermore National Laboratory in 2005 will have over 64,000 nodes. Groundbreaking architecture Blue Gene/L's architecture is unique, but the machine was designed to be familiar and easy to use as well. Architecturally, Blue Gene/L contains thousands of power-efficient processing chips, each with dual Power PC processor cores, on-chip memory, and two dual floating-point units to speed calculation. The system is integrated via multiple interconnection networks and compressed into a very dense package. This represents a new way of building a supercomputer. Yet while Blue Gene/L is not a Linux cluster, in many ways it looks and feels like one-quite deliberately. "We run the machine under control of the Linux operating system and we have designed it to ensure that programs written for clusters will run on Blue Gene/L," Bhanot said. While Blue Gene's speed and computing power open enormous possibilities, the transition for a user from traditional supercomputers to the Blue Gene/L environment should be easy for most users. Besides rewriting the rulebook for supercomputing, IBM's 19/04/2005 11:56
IBM Think Research | The Blue Gene revolution
6 sur 6
http://www.research.ibm.com/thinkresearch/pages/2004/20041214_bl...
architectural advances for Blue Gene/L will also lead to improvements in the company's mainstream product portfolio. All the knowledge gleaned from overcoming the hardware, software and application challenges inherent in building this ultra-powerful system will be leveraged to develop new products and solutions across IBM. The road to a petaflop As the Blue Gene/L project progresses and IBM continues scaling up to bigger machines, the company is exploring ways to create a community of Blue Gene/L users at the university/research level. IBM believes the fastest way to truly test Blue Gene/L's capabilities is to allow the scientific community to tackle its toughest challenges on this new architecture. In addition to the Lawrence Livermore and ASTRON machines, IBM is building a Blue Gene/L supercomputer for AIST, a major Japanese Computational Biology Research Center. IBM's Computing On Demand centers will provide a way for IBM clients to work on Blue Gene/L systems on a pay-for-use basis. Blue Gene/L is the first in a series of supercomputers under development that will push the limits of high performance architecture and computing power. Plans are already underway to develop a successor to Blue Gene/L that will reach the one petaflop (one thousand teraflop/s) performance level-and beyond that, IBM's sights are firmly set on building a machine capable of performing at several petaflops. IBM also recently announced the commercial availability of IBM eServer Blue Gene Solution, a full rack system that will deliver a peak performance of 5.7 teraflops. Based on IBM's Power architecture, the IBM eServer Blue Gene Solution is optimized for bandwidth, scalability and the ability to handle large amounts of data while consuming a fraction of the power and floor space required by today's fastest systems. "I am confident that a ten petaflop machine is achievable by the end of the decade," Bhanot said. "Just a few years ago that number seemed inconceivable, but now it is realistically within our grasp." Richard Silberman is a freelance writer based in New York City ibm.com | IBM Research | About Think Research | Terms of use | Privacy
19/04/2005 11:56
