UH-Hilo gets supercomputer
HILO » The Maui High Performance Computing Center has donated its "Huinalu" supercomputer, capable of 250 billion arithmetic calculations per second, to the University of Hawaii at Hilo.
When the computer was installed on Maui just five years ago, it was one of the 100 most powerful in the world, a university statement said.
"It is certainly a big thing for a school the size we have," said Judith Gersting, chairwoman of the school's Computer Science Department. UH-Hilo has just 3,442 students, most of them undergraduates, compared to 20,644 at Manoa.
Huinalu is known more formally as the IBM Netfinity Cluster, which indicates it is a cluster of "nodes," each comparable to a home computer, all linked in a single system.
There are 128 of the nodes, each with two .933 gigahertz Intel Pentium III processors, the university said.
Huinalu was used on Maui by the Department of Defense and researchers, but it is being replaced by one with 1,280 nodes, 10 times more powerful than the old one, Gersting said.
Huinalu was packed and shipped from Maui on Aug. 7, but UH-Hilo has not received it yet, Gersting said. Imagining 128 nodes, each about half the size of home computer, plus connecting cables, she estimated it takes up a good part of a large shipping container.
Exactly what UH-Hilo will do with Huinalu hasn't been decided, Gersting said. But the uses will fall into two areas, teaching students how to put together and use such a tool, and making it available for research in various sciences at UH-Hilo.
Three undergraduate students from Hilo have spent their summer on Maui learning how to manage Huinalu, Gersting said.
Hilo undergraduate students have also built their own mini-supercomputers by linking 10 to 15 personal computers. How to link components, called computer architecture, is the specialty of computer science professor Sevki Erdogan, who will be in charge of Huinalu on the Hilo campus.
Once linked, computers or nodes have to be instructed on how to work together to calculate answers to problems. They do that by each node working on part of a problem, and then the whole system combining 128 partial answers in the end.