eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
NASA is turning to IBM and its iDataPlex server array to build a new supercomputer that can execute 42 teraflops and will help collect data from a series of satellites that observe both the Earth and the universe.
On Sept. 23, IBM is planning to announce that it will build the new supercomputer at the NASA Center for Computational Science, in Greenbelt, Md. When complete, this cluster-style supercomputer based on the iDataPlex array will offer a top performance of 42 teraflops and use 1,024 quad-core Intel Xeon processors. NASA is then planning to combine this new system with the existing Discovery supercomputer at NCCS, which will then offer a combined performance of 67 teraflops.
In addition to collecting a range of data from observational satellites, the new supercomputer cluster is slated to help NASA with its climate and weather modeling as well as creating simulations to explain cosmic phenomena such as black holes.
IBM is known for its range of supercomputers, including the well-known Blue Gene systems and its recent Roadrunner system at Los Alamos National Laboratory in New Mexico, which broke the petaflop mark. With clusters built around the iDataPlex system, IBM uses more industry-standard hardware, including Intel processors and InfiniBand switches.
When IBM first announced iDataPlex, Big Blue presented the array as a way for companies to use x86-based hardware to build data centers that can support a cloud computing infrastructure. However, IBM has announced a series of contracts with companies that will put iDataPlex to use as a supercomputer.
In addition to NASA, the University of Toronto has contracted with IBM to build Canada’s most powerful supercomputer using iDataPlex, which will create a system that offers a performance of 360 teraflops. Microsoft is also using a system based on iDataPlex that will test its HPC (high-performance computer) operating system HPC Server 2008.
In the case of NASA, Herb Schultz, a marketing manager with IBM’s Deep Computing division, said the NASA requirements for this supercomputer are similar to the needs of those companies developing their own cloud infrastructure or building out a business based on Web 2.0 technology.
“If you look at HPC requirements and the emerging requirements of Web 2.0, whether it’s social networking or gaming, there are a lot of similarities and a lot of it has to do with being able to scale out in a cost-effective manner,” said Schultz. “They also need to manage their power and cooling constraints and they need something that is incrementally global and something that is standard and running Windows or Linux on an x86 platform … For the kind of work NASA needs, it’s not too different than what we first talked about when we introduced iDataPlex.”
At the same time, IBM had to take into account NASA’s requirements for power and cooling as well as fitting the computer into a constricted space.
“When it comes to supercomputers, it used to be performance, price performance and can you run my codes,” said Schultz. “Now, it’s basically customers are asking that the system works underneath a certain power consumption threshold or they ask what are your flops per watt and can I put this in here without rearranging my data center.”
With the iDataPlex cluster for NASA, IBM is offering a number of power and cooling technologies including the company’s Rear Door Heat Exchanger, which is a series of sealed tubes within the cabinet that are filled with chilled water that helps cool the heated air before it comes out of the servers. IBM also rotated the racks within a typical iDataPlex array cabinet, which creates an environment that is wide but not deep, and allows the company to squeeze more servers into the system while leaving room for switches.
The new NASA supercomputer will also take advantage of IBM’s xCAT (Extreme Cluster Administration Toolkit) management software and the company’s General Parallel File System, which will allow the cluster to create and maintain large file sets based on the amount of data the cluster can process.