NERSC’s David Skinner Gets Hands on IBM Blue Gene Prototype
April 1, 2004
David Skinner, who is the lead NERSC consultant for the Seaborg system, was working on a performance monitoring project at IBM’s Watson Center in New York earlier this year when an unexpected opportunity presented itself.
“There was a BlueGene/L prototype available while I was there, and IBM Research was interested in getting more experience with real users and codes on the machine,” Skinner said. “This was the first real contact NERSC has had with this architecture, which is one of the more novel architectures to come along in a while.”
Over the course of three days, Skinner was able to use about 12 hours of run time on the prototype, completing several benchmark runs, as well as a couple of applications. “This data gives NERSC a reasonable starting point for understanding BG/L as an HPC resource,” Skinner said. The results were presented at a recent LBNL and discussions about future work on BG/L are ongoing.
Skinner, who earned his Ph.D. in chemistry at UC Berkeley, primarily consults with NERSC users about improving the performance of their codes on Seaborg. As part of NERSC's Science-Driven Computing effort, Skinner is also involved in studying performance issues on new architectures such as Blue Planet, BlueGene and parallel vector systems. But he has been branching out into studying various architectures and performance issues.
It was this interest that led to informal conversations with IBM staff at scientific computing conferences. From this, Skinner spent a week or so in Yorktown Heights, working on a couple of projects related to hardware performance monitoring. One objective was to improve the scaling and performance of the parallel performance monitoring tools themselves. Performance monitoring that functions efficiently at high concurrency allows researchers to view the performance of an entire parallel calculation in a simple consistent picture. "The work done in January got us a five-fold improvement in profiling a 1024-way calculation," Skinner said.
In return, Simone Sbraglia of IBM will be spending two weeks at LBNL’s Oakland Scientific Facility beginning in late April. Sbraglia is the main developer of the Simulator Guided Memory Analyzer (SiGMA), a performance toolkit under development to help programmers understand the precise memory references in scientific programs. The goal for Sbraglia’s visit is to apply SiGMA to examples of the NERSC workload in order to get a better understanding of memory access patterns and performance in scientific codes.
It will also help shake out how well a tool like SiGMA can work on real world codes and provide IBM Research some indication about SiGMA’s use as more than a research tool.
“Visiting with developers is a great way to educate one another about what works and what doesn't,” Skinner said. “Case in point, NERSC has many more applications than most computing centers and we can't always dig down to the deepest levels for each code. Consequently, tools that can profile with little overhead are especially valuable. There's a lot of cross-pollination that can go on during these exchange visits.”
For more information, contact David Skinner at email@example.com.
About NERSC and Berkeley Lab
The National Energy Research Scientific Computing Center (NERSC) is a U.S. Department of Energy Office of Science User Facility that serves as the primary high-performance computing center for scientific research sponsored by the Office of Science. Located at Lawrence Berkeley National Laboratory, the NERSC Center serves more than 7,000 scientists at national laboratories and universities researching a wide range of problems in combustion, climate modeling, fusion energy, materials science, physics, chemistry, computational biology, and other disciplines. Berkeley Lab is a DOE national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California for the U.S. Department of Energy. »Learn more about computing sciences at Berkeley Lab.