May 2017

Future Online Analysis Platform Workshop

ICL’s Piotr Luszczek and Terry Moore participated in a workshop on April 4th and 5th in Washington, DC, directed at confronting edge computing environments, which are characterized by the rapid proliferation and deployment of new and powerful data sources, ranging from huge instruments and experimental facilities to highly dispersed sensor networks.

Called the Future Online Analysis Platform, the workshop was sponsored by the US Department of Energy (DOE) and organized by Kate Keahey of Argonne National Laboratory and Jim Ahrens of Sandia National Laboratories. The workshop was designed to contribute to DOE’s effort to develop scalable cyberinfrastructure that can provide efficient and timely data analysis capabilities to researchers who want to channel and use this unprecedented flood of data.

Piotr participated as a representative of a nascent project focused on scalable distributed computing for data analysis in edge environments. Known as InLocus, the project is being developed by ICL with Martin Swany of Indiana University and Micah Beck of UT Knoxville.

Terry participated as a representative of the Big Data and Extreme-Scale Computing (BDEC) project. BDEC is premised on the need to systematically map out and account for the ways in which the major issues associated with Big Data intersect with, impinge upon, and potentially change the national (and international) plans that are now being laid out for achieving exascale computing. ICL has been leading the BDEC project the past several years.

Invited Talk at NASA Langley Research Center

During a visit April 11–12 to NASA Langley Research Center in Langley, VA, ICL’s Stephen Wood gave a talk on the numerical analysis and linear algebra subject of LU (“lower upper”) factorization to approximately twenty-five members of the center’s Computational Aerosciences Branch.

After his talk, Stephen spent the rest of his stay improving the performance and scalability of a finite element mini app by refactoring OpenMP commands; attending a talk on LAVA by Cetin Kiris, the NASA Computational Aerosciences Branch chief; attending a talk on transition modeling by James Coder of UT Knoxville’s Mechanical, Aerospace, and Biomedical Engineering Department; and participating in discussions concerning the HPC needs and challenges across NASA’s branches.

LAVA is a computational framework for launch, ascent, and vehicle aerodynamics.

Transition modeling pertains to predicting the changes between laminar (streamline) and turbulence flows in fluids and the effects on the overall solution. Transition modeling is very compute intensive.

“Detecting and quantifying the transition from laminar flow to turbulence and back again is one of the major frontiers of simulation and experimental science,” Stephen said. “In many respects the aerospace community is leading the charge to better understand turbulence, but the impacts of progress are felt in disciplines from meteorology to cardiology.”

Invited Talk for the DoD High Performance Computing Modernization Program

On April 25th in Aberdeen, MD, ICL’s Stephen Wood gave an invited talk at the US Department of Defense High Performance Computing Modernization Program PETTT Transition to Knight’s Landing (KNL): Approaches, Tools, and Architecture workshop to an interactive audience of approximately sixty people in person and thirty web participants. PETTT refers to user productivity enhancement, training, and technology transfer.

“The presentation sparked fruitful discussions about the influence of work-sharing patterns, the utility of OpenMP features, and the configuration of computational environment—KNL memory modes, numactl settings, numeric libraries, and so forth,” Stephen said.

Lessons learned during the development of what is called an asynchronous parallel LU factorization (PariLU) for dense and sparse matrices on Haswell, Knights Corner, and KNL processors were presented and discussed, along with other related subject matter.