ATPESC gives participants a crash course in supercomputing

science
ATPESC 2015
Pooja Rao
ATPESC 2015
Bill Gropp works with ATPESC attendees

For two weeks this summer, a group of 65 students and early career researchers took up residence at Pheasant Run Resort’s Gallery Hall in St. Charles, IL for an arduous training program designed to teach them the key skills and tools needed to efficiently use leading-edge supercomputers.

Packed with technical lectures, hands-on exercises and dinner talks, the 2015 Argonne Training Program on Extreme-Scale Computing (ATPESC) featured an intensive agenda that began with presentations at 8:30 a.m. and ended with optional collaborative sessions that often lasted until 10 p.m.

“I know of no other program that offers such a comprehensive and in-depth training in high-performance computing,” said ATPESC attendee Jason Bender, a PhD candidate at the University of Minnesota and a fellow in the DOE Computational Science Graduate Fellowship program. “The program is definitely intense, but it is its total immersion approach that makes it so effective.”

Paul Messina, Director of Science at the Argonne Leadership Computing Facility (ALCF) conceived and developed ATPESC as a way to grow the high-performance computing (HPC) user community by filling a gap in the training most computational scientists receive early in their careers. The program, which began in 2013, is funded by the DOE Office of Science. ALCF is a DOE Office of Science User Facility.

The program addresses all aspects of high-performance computing with a curriculum that evolves each year to emphasize particular areas of interest. This year, the organizers incorporated more hands-on sessions and placed increased focus on the importance of performance portability across diverse computing architectures. The content was organized around seven core program tracks:

  • Hardware Architectures
  • Programming Models and Languages
  • Numerical Algorithms and FASTMath
  • Community Codes/Software Engineering
  • Visualization and Data Analysis
  • Toolkits and Frameworks
  • Data Intensive Computing and I/O

In addition, participants were provided access to hundreds of thousands of cores of computing power on some of today’s most powerful supercomputing resources, including the ALCF’s Mira and Vesta systems, the Oak Ridge Leadership Computing Facility’s Titan system, and the National Energy Research Scientific Computing Center’s Edison system. Titan and Mira are the second and fifth fastest supercomputers in the world, respectively.

For Ivy Bo Peng, a PhD candidate at KTH Royal Institute of Technology in Sweden, this unique opportunity allowed her to use Mira to test iPIC3D, a plasma physics code she is working on. To determine the problem size that can be run efficiently with the code, she ran a weak scaling test on 32 racks of Mira (524,288 cores) that achieved 81 percent efficiency.

Peng plans to continue her code development work at the ALCF in the future by applying for a Director’s Discretionary award.

“ATPESC has given me a bigger picture of HPC and extreme-scale computing,” she said. “My research may only focus on one small area, but I will now be paying more attention to the interplay with other aspects of HPC.”

Tim Warburton, a professor at Virginia Tech, presented lectures on the programming of accelerators, including graphics processing units. He believes ATPESC is an important endeavor that can help address the shortage of researchers who are capable of advancing state-of-the-art algorithms for use on leadership-class supercomputers.

“The program provides junior researchers insight into the possibilities that these systems offer at the critical point in their training, when they can steer their nascent research efforts towards scalable high-performance computing,” Warburton said. “This year’s class offered a very encouraging snapshot of the next generation of HPC-savvy researchers who are able to marry advanced computing techniques with advanced numerical methods.”

Selected from a highly competitive field of 170 applicants, program participants were comprised of doctoral students, postdocs, and computational scientists who have used at least one high-performance computing system for a reasonably complex application and are engaged in or planning to conduct research on large-scale computers. Their research interests span the disciplines that benefit from supercomputers, such as physics, chemistry, materials science, computational fluid dynamics, climate modeling, and biology.

For Eduardo Sanchez, the program offered a nice complement to his recently completed PhD in computational science at San Diego State University.

“I would make ATPESC mandatory as part of the completion of any PhD program in the field,” he said. “It will surely make me consider taking on more HPC-intensive projects in the future.”

For more information on ATPESC, visit http://extremecomputingtraining.anl.gov.