IBM puts Blue Gene to work on Big Bang

Discussion in 'Philosophy' started by Superjoint, Feb 23, 2004.

  1. Dutch astronomy lab to use IBM supercomputer

    By Stacy Cowley, IDG News Service February 23, 2004



    IBM Corp. plans to put some of its big iron to work next year to help a Dutch astronomy lab construct a radio telescope that scientists hope will let them gaze deeply enough into space to study the moments surrounding the universe's creation.

    The Netherlands Foundation for Research in Astronomy (Astron), in Dwingeloo, has been working for several years toward a radio telescope project known as Lofar. Lofar, which stands for "Low Frequency Array," will use thousands of radio antennas spread over several hundred miles to detect cosmic signals -- including those from objects so distant their radio signals are believed to have been emitted just after the Big Bang.

    "The big problem with existing telescopes is that they only look in one direction at a time," said Harvey Butcher, Astron's executive director. "Lofar will be able to see the whole sky."

    Logging data from scanning the whole sky takes dramatic processing power and throughput, which is where Big Blue steps in. Because deep astronomical research usually requires supercomputers more advanced than any available commercially, scientists are accustomed to building their own machines. But, in a bid to control its costs, Astron decided more than a year ago to look for a commercial partner for Lofar's computing needs. IBM responded with a pitch for Blue Gene, the family of supercomputers IBM intends to be the world's most advanced.

    In the works for several years, IBM's first wave of Blue Gene/L machines are scheduled for release in mid-2005. The series' high-profile patriarch will be a system ordered by the Lawrence Livermore National Laboratory, featuring 64 refrigerator-sized racks capable of processing more than 350 teraflops per second -- not far off the total computing power of the top 500 supercomputers in the world today.

    In terms of processing power, Astron's Blue Gene/L will be the baby of the family, with six racks. That's still enough to lodge Lofar on the top-10 list of the world's supercomputers. And it's in bandwidth that Lofar will surpass its siblings, according to Bill Pulleyblank, IBM Research's director of exploratory server systems.

    "We took the same type of Blue Gene we're building for the Livermore lab and we reconfigured it to get a lot more I/O (input/output) bandwidth into it," Pulleyblank said. "That was the design challenge. Although it's much smaller in the number of processors, we increased its data-handling capacity."

    Lofar will have as much as 20 terabytes of data flowing through its pipes each second, he said. That's enough data to encode more than 3,000 music CDs.

    Astron's Butcher said his group's goals are right at the edge of what modern computing technology permits.

    "There's a whole period in the early universe where people don't know what happened. We're hoping that Lofar will be the first telescope that will be able to see the first objects being formed. Maybe, with a bit of luck, we'll see what they are," he said.

    Astron aims to have Lofar running by 2006 -- a deadline imposed not by scientists or bureaucracy but by nature. Radiation from the sun disturbs the incoming radio waves Lofar will detect. In 2006, the sun will be at a low ebb in its 11-year activity cycle, a window Butcher said Astron hopes to exploit so it can work the kinks out of Lofar during a quiet period.

    Astronomy will be the project's main focus, but other scientists may queue up to take advantage of the system's processing and data-gathering resources. Butcher envisions attaching other devices -- such as seismic, wind, and weather sensors -- to Lofar's antennas, allowing researchers in other fields to pursue their own initiatives.

    "We see this project as a technology platform that is much broader than just astronomy," Butcher said. "It should be very exciting."
     
  2. woooaaah.
     
  3. 20 terabytes a second is freakin fast as f*ck!!!!
    that is just crazy



    and how fast is a teraflop?
    I'm guessing its like
    1000000000000000000000 Mhz or something
     
  4. A teraflop is a trillion floating point operations. There is no conversion to processor frequency since flops are purely arithmetic, and computational dynamics are rarely purely arithmetic.

    For record though, Blue Gene has more than 1200 Power PC processors, each is probably on the order of 5 to 6GHz. This, in itself, in not that impressive, as its predecessors have had more, and faster. The impressive piece is that it is scalable to over 4000 processors.

    To compare, a RS/6000 (i.e. Deep Blue) can take up to 512 processors on cards (each holding 8 processors) and 128GB of system memory (RAM). Blue Gene must have a comparable amount of system memory, say roughly 4TB.

    As for 20TB/s data throughput. That's not a particularly impressive number seeing at the RS/6000 is capable of handling 16 fiber cards, each with 16 fiber optic channels. That's a data input of roughly 2TB/s, and that says nothing about output, or internal processing. I'm sure the reason they quote 20TB/s is because that's all that Lofar can put out (or all it needs to put out).

    More impressive numbers have been posted by some of IBMs older equipment. Like its tape libraries. I saw a tape library at a Liberty Mutual Region Computing Center that had 16 frames each with 6 tape drives. Two robots worked 24 hours a day loading and unloading 200GB tape catridges at a rate of about 1 every 5 seconds. That is a raw data write speed of 40GB/s per tape drive, which doens't sound too impressive, except that there are no gimmicks. Its actual write speed. The library takes in about 50TB/s which it sends to a mainframe for compression, which sends it back to the library. If your home computer could write/read 40GB per second, you computer would boot before you took your finger off the power button. You could burn 500 songs per second onto a CD. Now multiply that by 92. Cost? Roughly $15 billion US.

    That's just storage though. Processing power is much cooler. At a cost of nearly a million dollars a processor, and more than a million per gigabyte of memory, Blue Gene may not be as expensive as a tape library, but it sure sounds more badass.

    Hope someone found this interesting.

    -XaWN
     

Share This Page