The usual hum in the basement of Yale Information Technology Services will soon be accompanied by the roar of a new high-performance computing cluster, called Beowulf.
ITS Director of Academic Media and Technology Charles Powell said research at the Yale School of Medicine is becoming increasingly computer-dependent. Beowulf will primarily be used by the medical school to search for patterns in fMRI digital images and genomes. These operations had required a lifetime to complete but can now be executed “by throwing tons and tons of computing power at it,” Powell said. The new cluster will be operational in a matter of months.
The costs for the new cluster are covered by a $1.5 million grant from the National Institutes of Health. Obtaining funding was difficult, Powell said, but the number of potential applications for a powerful Beowulf cluster is growing. He said he is confident Beowulf will be worth its hefty price tag, and once the University has its own advanced computing facility, faculty will be able to tackle complex problems more easily.
A group of 22 faculty members from the medical school and Computer Science Department organized the effort to secure NIH funding. After more than a year of planning, Powell said boxes of equipment started arriving three weeks ago. It will take at least a month to get the machine up and running, Nicholas Carriero, a research scientist in computer science, said.
A cluster is a collection of independent computers combined into one unified system — Beowulf is a specific type of cluster. Yale’s Beowulf will consist of 130 computers, providing 260 central processing units. Dell manufactured the parts and will also assemble the cluster in the basement of the main ITS building at 175 Whitney Ave. Once the cluster is up and running, users will be able to access Beowulf remotely from any location.
Although the programs needed for analyzing data with the cluster already exist, they must be reformatted to work on the new system, Carriero said.
Until Beowulf is operational, scientists at the medical school will continue to use smaller and slower clusters at their labs or borrow computing time from their collaborators at other institutes.
Beowulf will share basement space with machines that coordinate campus e-mail, payroll and student information. The room is equipped with auxiliary power supplies, a cooling system to avoid overheating and robots that constantly record data for backup.
A computing rule of thumb dictates that processor speed doubles every 18 months, but Carriero said that rule is no longer valid. He said the Beowulf will probably be sufficiently “new” for four to five years.