At the supercomputing conference SC14, held in November, Fermilab astrophysics and computing experts achieved a milestone with a demonstration run of the analysis framework software they are developing for the Dark Energy Science Collaboration (DESC) of the Large Synoptic Survey Telescope.
The LSST, whose construction is led by SLAC National Accelerator Laboratory, is currently in the advanced design phase and will be placed in Cerro Pachon, Chile. It will be the tool for the world’s largest imaging survey, taking repeated images of the southern sky beginning in 2020.
The software and data processing demands for the DESC are challenging, to say the least.
“LSST will truly be a next-generation survey: It will surpass preceding surveys in terms of data size in its first few months of operation,” said University of Pennsylvania astrophysicist Bhuvnesh Jain, spokesperson for the Dark Energy Science Collaboration.
More than 200 scientists from five countries are currently involved with DESC, and Jain expects the number of scientists involved in DESC to double in the next decade.
Fermilab astrophysicist Scott Dodelson, convener of the DESC Software Working Group, says the group is designing a framework for all DESC scientists that will facilitate their collaboration and use of tools built by the LSST project team. Steve Kent, Jim Kowalkowski, Marc Paterno and Saba Sehrish, all in Fermilab’s Scientific Computing Division, worked to develop the DESC framework. Sehrish ran the November demonstration.
The framework links some programs specifically produced for LSST with others written externally or by the scientists themselves. It runs them on supercomputers, networks such as FermiGrid and local resources.
“The scientists running the DESC workflows will not have to worry about details such as file transport or access to supercomputers to do their dark energy science,” Paterno said. “We demonstrated how this could be done.”
Dodelson and Kent say that the demonstration was very successful and bodes well for the DESC.
“The demo was an end-to-end simulation of LSST data and science analyses — that was the really important thing,” Kent said. “It was a walking-through of all the steps and with an eye on eventually expanding to the LSST scale.”
The group ran simulated images through the interlinked software until at the end it was run through CosmoSIS, a cosmological parameter estimation program to which Dodelson, Kowalkowski, Paterno and Sehrish have contributed.
The DESC Software Working Group is currently developing another version of the demonstrated framework for the DESC scientists to consider at their February gathering at SLAC.
Jain said that innovative DESC software will enable explorations of the many astronomical mysteries that LSST will open up.
“The work of the Fermilab group is really going to pave the way for a new mode of doing software analyses and how people collaborate,” Jain said. “I think it will have far-reaching implications for how we do cosmology.”