Research scientists Dr. Mehmet Belgin and Semir Sarajlic from the Partnership for an Advanced Computing Environment (PACE), in collaboration with Center for Relativistic Astrophysics (CRA) faculty Drs. Laura Cadonati, Nepomuk Otte, and Ignacio Taboada, received a $400k National Science Foundation (NSF) grant (award OAC-1925541) to extend the Open Science Grid (OSG) service to the Georgia Tech campus. OSG is an NSF-funded national computational grid that provides shared resources to run massive numbers of short/small computations, commonly referred to as High Throughput Computing (HTC). Since 2004, PACE has been successfully providing federated services for long-running, large-scale computations, known as High Performance Computing (HPC). In recent years, the number of HTC-type users utilizing PACE has grown rapidly relative to the number of conventional HPC-type users. As a first step towards supporting HTC-type workloads, PACE deployed a custom OSG cluster for the Laser Interferometer Gravitational-Wave Observatory (LIGO) project led by Cadonati. Peter Couvares (Caltech), chair of Computing for LIGO, said, “Georgia Tech’s collaboration with the Open Science Grid has been of tremendous benefit to LIGO, and I’m looking forward to PACE’s growing capabilities.” With this grant, PACE will be able to extend OSG support to the rest of the researchers on campus, which will improve the overall efficiency and utilization of campus resources. The three main features of the proposal include:
- Providing a centralized OSG support structure to Georgia Tech researchers, who can benefit from HTC, with an option to invest their research funding directly into OSG;
- Providing new GPU resources that are currently in high demand in OSG for large-scale, highly-parallelized computing; and
- Growing OSG’s data federation with the addition of the first “StashCache” service in the southeastern United States to enable fast, convenient access to large data sets.
“We are excited about the significant contributions this grant will add to GT’s existing capabilities,” said PI Belgin. “Perhaps more importantly, it marks the beginnings of a centralized PACE service to support very large numbers of relatively small and short computations -- a common type of computation that has been on the rise at GT.”
The immediate benefit of this grant will be to the multi-messenger physics programs supported through CRA faculty research. The augmentation of the existing LIGO cluster with additional resources brings new capabilities for Cadonati, who will be able to extend her analysis work and test the new GPU-based LIGO software stack. Cadonati explained, “This cluster is placing Georgia Tech at the forefront of the quest for understanding the mysteries of the universe, analyzing data from gravitational waves, photons and neutrinos; we are now pushing the frontiers of science!” Also benefiting from these GPU nodes will be Taboada of the IceCube collaboration, which currently consumes more than 99% of OSG’s GPU resources. Otte, whose work with the Cherenkov Telescope Array (CTA) and the Very Energetic Radiation Imaging Telescope Array System (VERITAS) entails hundreds of terabytes of data, will benefit from a StashCache instance local to Georgia Tech. CTA computations have been run in Europe, mostly due to the lack of a high-capacity StashCache server to store large sets of CTA simulation productions. Those computations will now be able to run in the US.
This NSF award will profoundly impact scientific research. The addition of computational resources locally provides more than just an increase in hardware; it complements the existing HPC infrastructure at PACE with a new HTC-focused model and provides increased opportunity for all researchers on campus and nationwide that participate in the OSG.