This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
The Cosmic Microwave Background (CMB) is the earliest possible image of the Universe, consisting of primordial photons last scattered only 400,000 years after the Big Bang. The statistics of the tiny anisotropies in the CMB temperature and polarization encode not only the parameters of cosmology, but also insights into fundamental physics at the highest energies. In addition CMB science has been very successful in engaging the public. Since its discovery, CMB experiments have sought to detect ever fainter signals at ever higher resolution, requiring ever increasing data volumes. For the last decade the analysis of CMB data sets has become a high performance computing challenge. This project will take the current state-of-the-art CMB simulation and analysis code suite and undertake the computer science research program needed to take it from the tera- to the peta-scale and beyond, informed by the actual requirements of some of the most computationally-challenging planned CMB polarization experiments, including the proposed CMBpol satellite mission.
The project will start by creating a stripped-down test code that retains all of the computational but none of the scientific complexity of the existing core code base, which will be used to implement a range of possible solutions to known current bottlenecks. The second phase will pass the lessons learned with the test code back into the full science code suite, and use it to perform previously intractable analyses as proof of performance. This includes implementing run-time self-tuning. The project will also use its proof of performance tests to undertake currently intractable analyses in support of next-generation experiments, to field a new tool for application-driven whole-system performance analysis, and to introduce a new generation of data analysts to ultra high performance computing.
The research team includes PI Simon, an authority on high performance computing systems and trends, as well as sparse matrix algorithms, and CoPI Borrill, who has led the efforts to deploy high performance computing for CMB data analysis, both on specific missions and for the international CMB community as a whole. Long-term collaborator Stompor works with the Planck HFI data analysis team in Paris, and faculty associates Yelick and Demmel add valuable expertise. Leading scientists associated with many of the current ground-based, balloon, and satellite CMB experiments have agreed to lend their expertise, as associates to this project.
