POWER > Trends > IBM Research

DOME-inating Research Looks at the Big Bang


IBM scientist Ronald Luijten works on an exascale computer as part of a five-year collaboration with ASTRON, the Netherlands Institute for Radio Astronomy. Photo courtesy of IBM Research - Zurich

Imagine being able to look into the night sky and see actual remnants of the Big Bang. If a telescope could peek back that far in time, it might be possible. But until—and if—that ever actually happens, scientists have other ways to capture the essence of the universe-creating event, including observing radio waves.

Unfortunately, current radio-wave observatories aren’t capable of capturing all of the radio information out there, and even if they could, processing and storing all of that data would be a daunting task. To reach this point, ASTRON—the Netherlands Institute for Radio Astronomy—and IBM are developing a technology roadmap as part of an initial five-year technological collaboration, dubbed DOME, on the Square Kilometre Array (SKA) in Australia and South Africa.

Using cutting-edge technology, SKA researchers may be able to listen to the universe as it burst into being some 13 billion years ago, as some scientists estimate. And thanks to IBM researchers such as Ronald Luijten, IBM data motion architect, they could potentially do this in a data-rich and processing-intensive environment that will focus on, among other things, energy conservation.

Q. Before we get to SKA, why are you interested in data motion?
A. We have a huge challenge coming toward us in the next five years or so to build data centers that are very energy-efficient yet need to deal with huge amounts of data, or what IBM likes to call big data. But as it turns out, 98 percent of the energy consumed in our data centers is used for moving data from point A to point B, whether it’s from the hard drives to our servers, or our memory to our caches or our caches to our compute cores. Only 2 percent is actually done in what banks would consider the value-add operation of adding two numbers, for instance. In fact, I call data motion a sin.

Q. A sin because it consumes so much energy?
A. Exactly. For the first chip I designed at IBM, we had a few gates only—maybe 2,000—and many hundreds of pins. The gate was the most expensive part and we didn’t even consider the pin cost. As a result, in those days, you’d have very few compute operations and would move data around like there’s no tomorrow. Currently, we’re still trying to come to grips with needing to optimize very differently based on the technology changes that occurred over the past three decades. This is why I call data motion a sin.

Q. How does the SKA project fit into this thinking?
A. First, let me give you some background on SKA, which is really about trying to understand what we are and why matter is here today. What they’re going to study is what happened at the time of the Big Bang some 13 billion years ago. Believe it or not, it’s actually possible to measure what happened at the time of the Big Bang, because the radio waves that were generated at that point are still travelling through space at the speed of light. If you want to measure those radio waves, you need to look very, very far out into space.

To do so, you need to build a radio antenna that’s extremely sensitive, and the only way you can make such an antenna is to make it very large. The SKA is literally a square kilometer array, and what we’re building is a single machine—a single instrument that has a total collecting area of 1 square kilometer, or 1 million square meters. If you were to build this with your TV satellite dishes that you use for your home, you would need 3 million of them, and those 3 million would be part of one single instrument. Now, SKA scientists aren’t going to use those dishes to build their machine; they use a combination of 15-meter dishes and flat antenna “aperture arrays.” In total, they will need in the order of several thousand of these different types of antennas that are part of the single machine.

However, the challenge to building is that those antennas are going to be distributed across a very large baseline. Just recently, the SKA governing body announced that the arrays will be located in South Africa and Australia.

What excites me about the project are mainly the technical challenges, including that this antenna is going to generate 1 exabyte of data per day that we want to store. One exabyte is twice as much as what is being sent through the Internet on the planet in a single day. We’re talking big data to the extreme. The word “big,” as in “big data,” is simply nowhere near describing what the volume of this data is.

You might ask why IBM is all of a sudden interested in radio astronomy. It’s nice in itself, but how does it help IBM with its problems? Well, the key message is, number one, big data. The entire world, the commercial world, the way people keep track of what they’re doing themselves, the pictures they generate, the social networking—all of this is turning into big data, and IBM is tying all of the various elements of data together to better manage the planet. So the number one synergy we have with radio astronomy is big data.

The second one is that the amount of processing that needs to happen to deal with this amount of data from the antenna is in the order of exascale. IBM built the first petaflop machine called Roadrunner at Los Alamos in 2008, but we want to get a thousand-fold increase in performance every 10 years. In fact, the supercomputing guys are expecting to get 1,000 times over a petaflop, which is an exaflop, by 2018.

But regarding both SKA and exascale, currently we don’t know how to build those machines in a commercially viable way. If we were to build an exascale computer today, it would use so much power that we couldn’t afford to let them run more than 10 minutes a day because of the electricity bill.

Now, there are systems such as LOFAR, a low frequency array that has been just completed in the Netherlands. It’s a 1 percent demonstrator of SKA, so we know that the science works. The challenge with SKA is really one of scaling to something that’s 100 times larger. We can’t afford to use 100 times as much energy. So we need new ways of figuring out how we can build this machine in an acceptable way.

IBM and ASTRON got together about two years ago and we discussed about 20 different topics that relate to the needs of SKA and IBM. Out of that meeting, we selected seven projects of strategic importance to IBM for its commercial work as well as for the needs to build the SKA machine. We wrote a proposal to get funding, which we did through the Dutch government.

Jim Utsler, IBM Systems Magazine senior writer, has been covering the technology field for more than a decade. Jim can be reached at jjutsler@provide.net.


comments powered by Disqus

Advertisement

Advertisement

2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store