Argonne's Superstar Supercomputer
Mira is a superstar or, more precisely, a supercomputer that will be arriving at Argonne National Laboratory. The new technology may help build safer cars, develop new medical therapies, advance airplane design and create better climate models, according to Michael Papka, deputy associate laboratory director for Computing, Environment and Life Sciences at Argonne. Researchers from around the world, including those at Northwestern University and the University of Chicago, can apply to use Mira’s mighty processing power for their research.
Why should the average person be excited about a new supercomputer?
Mira will help us produce more accurate models of everything from cars to the human body. The more accurate your models are, the better the information you have to make decisions. This new supercomputer means we can potentially improve the efficiency and manufacturability of airplanes, develop cars that burn less gas and provide more passenger safety, and find more effective solutions to health issues. Everything is driving to improve the quality of life and wellbeing of the world. It seems pretty grandiose, but it’s really why Mira and its predecessor, Intrepid, are allocating time to the international community.
In the past, designers would make a full-size model of the car with clay and then put that model in a wind tunnel to get measurements. If you’re building a physical model, you may only have the money to try five ideas, even if there are other ideas that might be better or safer. As we get more accurate computer simulations, we can run many tests before having to build physical models. It saves time and allows more possibilities to be explored, leading to better designs. In designing the 787, its latest airplane, Boeing needed to build only seven prototype wing designs. For its prior-generation airplane, the 777, Boeing built and tested 77 wing prototypes.
What will this computer allow us to do that computers have not been able to do before?
We will be able to tackle larger problems and achieve faster times to solutions. For instance, with weather forecasting right now, models are on a multi-kilometer grid, which means meteorologists can predict rainfall within a two-kilometersquare area. The next generation machine will allow us to model that with more detail, so that rainfall predictions can be made within a half-kilometer square. This could be important information for people living in areas that are vulnerable to flooding.
How did you choose the name Mira for the new supercomputer?
Argonne's Blue Gene/Q computer system was named Mira as a reference to the Laboratory's approach to scientific computing activities. Historically, Argonne has often engaged in groundbreaking computing activities that were initially considered strange and, as they matured, inspired wonder. Mira is a Latin root for wonderful, strange, remarkable, amazing, surprising, and extraordinary. In addition, a variation of "mira”—mirino—means viewfinder, and is a reference to the path-finding nature of the BG/Q.
Mira is a 10-petaflops supercomputer. How fast exactly does Mira think?
A flop, short for floating-point operations per second, is a performance measurement describing the number of arithmetic operations per second that a machine can do. You can compare Mira to an iPad to understand its capabilities. Mira will be approximately 10 petaflops (10,000,000,000,000,000 flops) and an iPad is only 171 megaflops (171,000,000 flops). It would take more than 58 million iPads to equal Mira’s computing power. It would still not equal Mira’s integrated power, however, with its many special features.
How much more powerful is Mira than its predecessor? Is it the fastest computer in the world?
Mira is 20 times more powerful than Intrepid, Argonne's current supercomputer. The number one computer right now on the TOP500 List of the world's most powerful supercomputers is the Japanese K machine. Japan and China are really pushing things forward. However, at the same time IBM is deploying Mira to us, it is also deploying a machine twice as big to the Lawrence Livermore National Laboratory—a 20 petaflops machine called Sequoia. On the November TOP500 list, Sequoia could be No. 1 if China and Japan don’t do anything.
When do you expect Mira to be up and running?
Our hope is that we will receive the machine sometime in the third quarter of 2012 or early in 2013. Then it will take us three or four months to stand it up. It’s made up of 48 racks that weigh 2 tons each, so it takes a while to wheel it in, put it in place and wire it up.
What does the new supercomputer look like?
I’m always surprised when people ask to tour our machine room because it’s a lot of noisy, boring space. There are no gears turning, it’s just rows of big computers. If you opened one up, you would see a bunch of drawers. If you pull the drawers out, you see a bunch of processors. There hasn’t been a radical change in appearance from the last supercomputer. Mira will be three long rows of big, black boxes.
Private companies, universities and other researchers all need time on supercomputers. How competitive is the process for corporations and researchers to get time on the supercomputers?
Allocations on the two Department of Energy leadership computing machines—Intrepid at Argonne and Jaguar at Oak Ridge—totaled 2.7 billion hours of computing time in 2011; however, average total requests from around the world are for three times that amount. We are limited in the numbers of hours we can provide, which means even good projects don’t make the cut. The decision on who gets time is made in a two-stage review process. The final decisions are made by the directors of the two centers, Jim Hack and myself, with input from leadership at each center. Every investigator who is given time on these super computers is among the best in his or her field or scientific domain.
What are some cool projects being run on the supercomputer right now?
There is a current study on concrete that is pretty exciting. Concrete production is a $100 billion a year industry in the U.S. and generates a lot of carbon dioxide. A researcher from the National Institute of Standards and Technology is using the computer to figure out how to design better concrete that produces less out-gassing.
Will results from the supercomputer calculations affect industry and the products that we see as consumers?
We are starting to see commercial companies we talk to about the machine say, “That’s directly applicable to our work." We want to make applications, like the concrete project, known to our commercial partners so there’s a faster transition from basic science to the applied efforts. Hopefully Mira will make that happen and help us figure out how to compress the time frame of bringing technology to people.
Have a question about this article? Submit it here to SiS. Answers to select questions will be posted on the site.