The Square Kilometre Array (SKA) is an international collaboration aimed at building the world’s largest radio telescope through an unprecedentedly large collection area (the titular square kilometer). Measuring as one of the largest scientific efforts in human history, the SKA will also produce enormous amounts of data, requiring the use of correspondingly large computational resources if the SKA’s ambitions are to be realized.
The 13-country effort is planning to begin construction in South Africa (which will host the high- and mid-frequency dishes) and Western Australia (which will host the low-frequency antennas) in 2021, giving researchers time to make computing decisions – and time for new, more powerful systems to arrive. On the other hand, it also leaves researchers trying to practice and prepare for handling the SKA’s data in the absence of real-world datasets. In a recent article, Elizabeth Rosenthal of Oak Ridge National Laboratory (ORNL) highlighted how researchers are tackling that problem.
“Generating such a vast amount of data with the antenna array simulator requires a lot of power and thousands of graphics processing units to work properly,” said Ruonan Wang, an ORNL software engineer. So researchers from ORNL the International Centre for Radio Astronomy Research (ICRAR, based in Australia) and the Shanghai Astronomical Observatory (SHAO) turned to the world’s most powerful publicly ranked supercomputer: Summit.
Located at the Oak Ridge Leadership Computing Facility, Summit’s 4,608 nodes (each with two IBM Power9 CPUs and six Nvidia Volta GPUs) together deliver 148 Linpack petaflops, making it the perfect candidate to simulate handling the kind of data output that the SKA is expected to produce. “The Summit supercomputer provided a unique opportunity to test a simple SKA dataflow at the scale we are expecting from the telescope array,” said Andreas Wicenec, ICRAR’s director of Data Intensive Astronomy. “Summit is probably the only computer in the world that can do this,” Wang added.
The researchers simulated the SKA-Low Array, which will span 131,072 antennas. To do this, they used the Adaptable IO System (ADIOS), an ORNL-developed open-source framework designed to accelerate high-performance data transfer for simulations. The team also scaled their simulator up to all of Summit’s compute nodes using an ICRAR-developed tool called the Data Activated Flow Graph Engine (DALiuGE).
“The scientific data group is dedicated to researching next-generation technology that can be developed and deployed for the most scientifically demanding applications on the world’s fastest computers,” Klasky said. “I am proud of all the hard work the ADIOS team and the SKA scientists have done with ICRAR, ORNL, and SHAO.”
“This was far more complex than a normal application,” said Wang, whose work on ADIOS has spanned six years. “The faster we can process data, the better we can understand the universe.”