Two Bulldogs sit in the basement of Watson Hall, growling and pondering the secrets of the universe.
Bulldog A and Bulldog B are the two clusters that make up the University’s new Shared High Performance Computing Facility, which should be functional by the end of the month, Yale ITS Director of Academic Media and Technology Chuck Powell said Thursday. Yale professors said Monday that they are excited about the possibilities Information Technology Services’ cluster could offer.
“I think it’s important for the University to do high performance computing,” astronomy professor Paolo Coppi said. “It’s where a lot of science is going on.”
The facility consists of two clusters in which numerous computing nodes are connected together to perform large calculations, Powell said. Each node contains two CPUs. One cluster — which Powell said is theoretically twice as fast as the other — is made up of 132 AMD Opteron CPUs connected to each other by Myrinet, a special technology used to connect clusters. The other uses 160 Intel Xeon CPUs joined by a gigabit Ethernet connection, Powell said.
Powell said the two clusters will probably place Yale in the top 500 computer facilities in the world.
The clusters’ combined cost will be in the hundreds of thousands of dollars, Powell said. He said the nodes themselves cost between $2,000 and $3,000 and Myrinet adds a cost of at least another $1,000 per node.
“It’s cheaper and more efficient than buying a mainframe, but by no means free,” Powell said.
To access the system, faculty members log on from outside locations and fill out a slip giving information about their task, including the specific number of CPUs and time required, Powell said.
“They drop it in there and essentially can go away and pick it up when it’s done running,” Powell said.
Applied physics professor Sohrab Ismail-Beigi said there are many advantages to Yale having its own high performance facility. When using other facilities, Ismail-Beigi said, scientists often have to wait in line for others to use the system first and then are only allocated a limited amount of computing time. Ismail-Beigi said researchers often need to run their programs many times and waste a significant portion of their time allocation doing debugging.
Molecular, Cellular and Developmental Biology Department chairman Michael Snyder said high performance computing could have “an enormous impact” on the life sciences. Snyder, who is currently working with human genome sequences, said biology requires an increasing amount of computationally intensive analysis.
“We really do need some incredibly awesome power to be able to move that kind of data around,” Snyder said.
The Yale clusters, Ismail-Beigi said, should offer a “huge amount of freedom,” since there will be many fewer users at the University than other facilities. Ismail-Beigi said users will know the facility’s managers and there will be less restrictions on allocation. In addition, the various professors using the system will know each other’s needs and be courteous, he said.
“The turnaround [on running programs] should be faster,” Ismail-Beigi said.
Coppi said high performance facilities allow scientists to create computer simulations of stars and other subjects.
“You press go and watch the galaxy evolve,” Coppi said.
Another high performance facility was installed in the School of Management last summer and implemented this fall, SOM Research Systems Administrator Derek Hunt said. Hunt said the facility consists of about 120 processors.
Hunt said he had not spoken with anyone in ITS about interaction with its cluster but said it was certainly a possibility.
[ydn-legacy-photo-inline id=”18454″ ]