June 28, 2011

Student Fashions Technique to Capture InfiniBand Processes

Nicole Hemsoth

Many of the TOP500 machines rely on InfiniBand to power their computations, but since this is still something of a new technology, enhancements and new developments to refine how it works are still in development.

An undergraduate student at Northeastern University has devised a new way for supercomputers making use of InfiniBand to grab data during the computational process and retain it to ensure against loss of progress if there are system problems.

According to Gene Cooperman, the student’s professor, InfiniBand is “behind some of the world’s largest computers, and yet the number of people who understand the internals of the technology is very small, largely because it is relatively new.” He says that no one has been able to restart an InfiniBand process midstream and the student’s developments might allow scientists to conduct their work more efficiently.

The student, junior Greg Kerr, has been selected to present an hour-long talk on his findings at Recon in Montreal, Canada, on the first day of the event. Cooperman feels this is an honor since it will give attendees the rest of the conference to discuss his student’s work.

Full story at Northeastern University