Supercomputer readies for action

Yale purchased the Bulldog O supercomputer last March for $2 million.
Yale purchased the Bulldog O supercomputer last March for $2 million. Photo by Zoe Gorman.

The fastest supercomputer Yale has ever owned is almost research ready.

Since Information Technology Services acquired Bulldog O — or “Omega” — last March from an undisclosed government agency, the machine has arrived to West Campus and been in a beta testing phase since August. Though a select group of researchers and ITS personnel are still testing Bulldog O’s functionality, Chuck Powell, associate chief information officer at Information Technology Services, said he hopes to open Bulldog O for general faculty use in two to four weeks.

Bulldog Omega, or “Bulldog O,” Yale’s newest supercomputer, has been undergoing rounds of beta-testing on West Campus since August; Yale acquired it for $2 million, approximately half of its estimated market price.
Bulldog Omega, or “Bulldog O,” Yale’s newest supercomputer, has been undergoing rounds of beta-testing on West Campus since August; Yale acquired it for $2 million, approximately half of its estimated market price.

“We’re in the late stages of testing on a select group of researchers,” Powell said. “We think we’ve discovered the last couple of bugs.”

Ranked 275th in the world for speed and the fastest supercomputer owned by any Ivy League institution, the machine processes large amounts of scientific data at high speeds in order to further large-scale research projects.

Powell said faculty interest in using Bulldog O’s computing abilities has been growing. Steven Girvin, deputy provost for science and technology, said about 50 professors have expressed interest in using Bulldog O so far.

Physics and astronomy professor Charles Baltay said last spring that he planned to continue his research using another supercomputer, but Powell said Baltay has since changed his mind and decided to compute with Bulldog O.

Baltay could not be reached for comment.

“People always want to use the latest, newest [supercomputer] that’s out,” David Frioni, ITS’s manager of high performance computing.

The machine processes 52.53 trillion calculations per second — but because interest in its high speed processing capabilities among faculty is great, not all jobs can be completed at once. Powell said a committee appointed by the Provost’s Office and chaired by engineering professor Sohrab Ismail-Beigi will decide on a fair process to allot use of the supercomputer’s 704 nodes — individual system units — to faculty.

Frioni said ITS has configured a “queueing system” based on a policy from the faculty committee. Researchers will send jobs to the supercomputer through an interface that can be run on their personal computers, and the queueing system determines which jobs Bulldog O will process when, according to the committee’s policy, he said.

The faculty committee is implementing and evaluating the effectiveness of the queueing policy using one of Yale’s older supercomputers, Bulldog L. Bulldog O processes at speeds 20 percent faster than L, Girvin said, and is 5.5 times as large.

For now, the system is perfectly democratic, Ismail-Beigi said, but this might change in the future depending on demand for the machine and urgency of research.

Now, each researcher gets 5 percent of the computer’s capacity. The program computes a score for each team that submits a job based on whether the team has exceeded its allotted capacity within an average time of two weeks.

A research team that has recently used more than 5 percent of the computer’s capacity can still process its jobs, but they will not be prioritized, Ismail-Beigi said, adding that the system also takes into account the size of the job waiting in the queue, so smaller jobs don’t have to wait for long jobs to finish.

“This very sophisticated software makes the queueing system that prioritizes the jobs and tries to keep the machine as close to fully busy as possible while giving everybody a fair share,” Girvin said.

One third of Bulldog O’s capacity will be used by three Yale professors — William Boos, Trude Storelvmo and Nadine Unger — researching climate change, Frioni said. Funding from the Geology and Geophysics Department, along with money from the three professors’ lab setup budgets, covered about one third of Bulldog O’s cost

Girvin said last spring that the supercomputer’s $2 million selling price was about half of its total estimated value.

Correction: September 23, 2011

An earlier version misstated the number of calculations Bulldog O can compute per second. Bulldog O can process at 52.53 trillion calculations per second. In addition, it is ranked 275th in the world for speed — not 146th — according to a report published last June.

Comments