The Tianhe-2 supercomputer may be number two in the TOP500 rankings, but it’s tops in radio telescope applications. An early version of the software that will analyze data from the Square Kilometer Array (SKA) is now running on the Chinese petaflopper. When completed, the software will act as the “brain” of the future SKA system, which will be the largest radio telescope every built.
Known as the SKA Science Data Processor, the software will process the enormous amount of raw data delivered by the array of telescopes that comprise the SKA and turn it into something that can be understood by mere mortals, or at least astronomers. When built, the system will not only be searching for signals of alien life, but also will be used to help elevate our understanding of the inner workings of the universe. Principal areas of interest include the study of star and galaxy formation and the phenomenon of gravitational waves.
The SKA system will be 50 times more sensitive and 10,000 times fasters than any other radio instrument on the planet. The reason it’s named the Square Kilometer Array, is that the thousands of radio antennas that comprise the SKA will cover a square kilometer of real estate across South Africa and Australia. A telescope of that size and power will generate about an exabyte of data every day and will require at least a hundred petaflops of computing power to process it.
Construction of the array is set to start 2018, with the first radio telescope observation expected in 2020. But the software is already in motion and its debut on the Tianhe-2 signals an important milestone for the data analytics part of the SKA project. According to Professor Tao An of Shanghai Astronomical Observatory in China, the software prototype was initially run of 500 compute Tianhe-2 nodes, which was soon extended to 1,000 nodes. The next step, says An, will be to increase that to 8,500 nodes, which is the approximate number that they expect to use on future supercomputer systems that will be used when the SKA radio telescope is installed.
The 8,500 nodes represent about half of the Tianhe-2’s 16,000 nodes and about half its 33 petaflops (Linpack) of processing power. To get to the 100-petaflop minimum threshold for SKA analytics will require nodes that are at six times as computationally dense. Since the Tianhe-2 currently relies on the now-dated Intel Xeon “Ivy Bridge” CPUs and Intel “Knights Corner” Xeon Phi coprocessors, it shouldn’t be too difficult to field such a system in the 2020 timeframe.
According to Tianhe-2 director Professor Yutong Lu, getting the initial software exercised on a real supercomputer was the first step. “The most important part is the co-design and co-optimization of SKA data processing software set and supercomputers such as Tianhe-2, [while] preparing for the faster computers in a few years from now,” he said.