Abstract: The DSA-2000 project is proposing the construction of a new radio telescope array with a number of antennas exceeding that of any other radio telescope, either planned or in existence. The design of the array depends on a number of technologies that are novel in the field of radio astronomy, not least of which is the implementation of a "radio camera", which produces calibrated images from the raw data collected by the antenna receivers in real time. For DSA-2000, the aggregate input data rate to the radio camera is approximately 43 Tb/s, and the output rate, 23 GB/s. Radio camera processing will use a dedicated cluster, which is expected to require 2800 GPU accelerators, located near the array center. In this talk, we will review the unique aspects of the DSA-2000 relative to previous interferometer arrays, and in particular the concept, design and implementation of the radio camera. The prototype radio camera software depends on Kokkos for its implementation -- we discuss the rationale for the decision to use Kokkos, provide examples of the use of Kokkos in the software, review the development experience so far, and provide an outlook on the future development of the radio camera project.
mpokorny
commented
11 months ago
Title: Implementing a radio camera using Kokkos
Abstract: The DSA-2000 project is proposing the construction of a new radio telescope array with a number of antennas exceeding that of any other radio telescope, either planned or in existence. The design of the array depends on a number of technologies that are novel in the field of radio astronomy, not least of which is the implementation of a "radio camera", which produces calibrated images from the raw data collected by the antenna receivers in real time. For DSA-2000, the aggregate input data rate to the radio camera is approximately 43 Tb/s, and the output rate, 23 GB/s. Radio camera processing will use a dedicated cluster, which is expected to require 2800 GPU accelerators, located near the array center. In this talk, we will review the unique aspects of the DSA-2000 relative to previous interferometer arrays, and in particular the concept, design and implementation of the radio camera. The prototype radio camera software depends on Kokkos for its implementation -- we discuss the rationale for the decision to use Kokkos, provide examples of the use of Kokkos in the software, review the development experience so far, and provide an outlook on the future development of the radio camera project.