Every year the who’s who in supercomputing meet in Germany at the International Supercomputing Conference (ISC), a leading conference and exhibition on high performance computing. This year’s gathering took place from June 17-20 at the Dresden International Congress Center.

Surrounded by the Center’s parkland, Science Writer Markus Henkel met Oliver Rettig, IBM’s program manager of Cell Blade hardware, to talk about the new QS22 Cell blade and its role in the petaflop Roadrunner supercomputer.
Markus Henkel: Mr. Rettig, you participated in the Roadrunner Project. What was your task and what was the challenge for the new number one TOP500 system?

Oliver Rettig: I was in charge of hardware development for the Cell blade product line. This blade is part of the new number one system. Now that the Roadrunner Project is finished, I am still responsible for the market launch of the QS22.

My entire team of 120 specialists contributed to making the seemingly impossible happen. One of the many challenges was to get nearly 7,000 blades logistically organized; a lot of them needed to be produced in Asia. Thus the technical challenge of realizing projects of this dimension is the time management and the logistics part. Integrating the software and hardware of the QS22 Cell blade was an outstanding accomplishment.

In addition to that, the QS22 is really new. Introducing something that new in a company the size of IBM is incredibly difficult. Such things usually take time to implement. Therefore we are very delighted to have the Roadrunner as a great showcase.

Another challenge for the technicians was the installation. For example, the wiring. For the 6,120 QS22 blades, we are talking about 91.2 kilometers of fiber optic cable. The installation of the cables alone can take a couple of weeks.

Henkel: IBM is one of the big players in HPC. Why?

Rettig: IBM makes use of its size effectively. To get the best people appropriate for a project out of the company’s 360,000 employees, you need very good networking, and therefore efficient communication. Due to the great number of outstanding minds at IBM, projects like the Roadrunner can be developed.

Henkel: More output, more energy efficiency, always on behalf of the climate: Isn’t that contradictory?

Rettig: You need to take two things into consideration here. One is the modeling of the climate, which inevitably needs more processing power. Certain things are now becoming possible due to the performance of a new generation of supercomputers. A lot of meteorologists only have a grid of 10 x 10 kilometers to model the climate. For this they include every single factor into their calculation. But, due to the occurrence of thunderstorms, for example, the forecast needs to be very precise. The granularity of the grid would have to be 2.5 kilometers.

A good example is a small climate project in Stuttgart, using two buildings with a canopied patio. In this glazed patio, which is about 100 meters long, a micro climate is being modeled on a small calculating platform. With adequate processing power, this can be done on a much larger scale.

Energy efficiency is another topic that has, in fact, been moving in line with the improvement of the climate model for quite a while. In the last couple of years we managed to get big systems into the Green500. We know what our [PowerXCell 8i] processor generation is capable of. Compared to standard processors, which are bigger and need much more power, the Cell processors are much more energy efficient. To me, it demonstrates that we are on the right path.

Henkel: Is there a race in supercomputing comparable to the competition between Microsoft and Google? For example, in Japan the next generation of 10 petaflop supercomputers is being developed.

Rettig: In what period of time?

Henkel: 2011.

Rettig: Of course! But we do not fight each other to the finish — like Microsoft and Google do. And companies are not preoccupied with supercomputing to that extent. The field is both too special and too small for that. It is more about the advancement of technology. For IBM, it is the challenge of reaching the next goal and being the first one to do so. Therefore, we try to develop the most efficient system for our clients.

Henkel: But that is what the others want too, isn’t it?

Rettig: That’s where certain factors are crucial — those that reflect the quality of the company: energy efficiency, software efficiency, the exact wishes of the client. That’s why the Roadrunner is also a hybrid processor. It has both Opteron and Cell processors. The client can choose from the two components to achieve the perfect mix.

Henkel: Are the enormous investments justifiable? After all, the computers are paid for by public money. How do the taxpayers benefit?

Rettig: The cost for a system, which is number one, is definitely justifiable. You are right, so far there has been a gap between supercomputers and people. But today there are systems available for less than 7,000 €. For that rather small amount of money, there are many application areas served by supercomputers.

Possible applications in the medical sector are very important. IBM works with the Fraunhofer ITWM in Kaiserslautern to research computer tomography. Most of the tomographic data is only presentable in two-dimensional diagrams. Using a Cell processor, the data can be converted into 3D pictures. That way you can display a three-dimensional beating heart, turn it around, zoom in, show cuts in veins and arteries, and even blank out tissues. To research heart diseases, it is worthwhile buying a supercomputer.

Henkel: What are your future expectations for QS22-based supercomputers?

Rettig: We are looking forward to developing a lot of new applications for the QS22. We have a lot of clients in the meteorological branch who are already asking for this new generation. Apart from that, there are also a lot of everyday tasks that need solutions. That’s what we are researching now.

Henkel: Mr. Rettig, thank you very much.

—–

About the Author

Markus Henkel is a geodesist, science writer and lives in Hamburg, Germany. He writes about supercomputing, environmental protection and clinical medicine. For more information, email him at [email protected] or visit the Web sites: http://laengsynt.de and http://netzwelt.de.