SCIENCE

Pervasive Technology Institute at Indiana University announced that Andrew Lumsdaine and Jeremiah Willcock of its Open Systems Lab (OSL) are part of an international team that developed the "Search" benchmark for the Graph 500 List. This benchmark evaluates a system's ability to traverse extremely large data structures.

Similar to the Top 500 List, which is used to measure trends of the world's fastest computers, the Graph 500 List uses benchmarks to compare the performance of supercomputing systems. The Top 500 List, however, does not measure a supercomputer's performance on data-intensive applications. Thus, in June 2010, Sandia National Laboratories began an effort to develop benchmarks that better model these types of applications.

OSL researchers developed a set of benchmarks as part of the Graph 500 steering committee and contributed to the reference implementation by creating a graph generator and two example implementations for distributed-memory computers and clusters. The team worked with IBM and Argonne National Laboratory (ANL) to benchmark "Intrepid," ANL's Blue Gene/P supercomputer, which placed first in the inaugural Graph 500 list. Jeremiah Willcock, in collaboration with Fabrizio Petrini from IBM, wrote the benchmark implementation used in that entry.

"I'm very pleased with Jeremiah's work on this project," said Andrew Lumsdaine, director of the Open Systems Lab. "Talking about the fastest supercomputer raises the question 'fastest at what?' The Linpack benchmark used for the Top 500 List focuses on raw computational power. But that's like comparing cars by their top speeds in a drag race. Many important problems, especially in informatics-related areas and life sciences, emphasize data access rather than sheer computation. The Graph 500 List is intended to test a system's abilities for these kinds of applications—it is more like comparing cars in a grand prix."