ITS finalizing new supercomputer queue

Following the University’s purchase of the “Bulldog O” supercomputer last spring, Information Technology Services is now working to equalize the access that Faculty of Arts and Sciences research groups have to it and the four other supercomputer clusters.

In response to increased faculty demand for access to Yale’s high performance computers, a provost-appointed subcommittee drafted and passed a policy earlier this academic year that streamlined guidelines for supercomputer usage, committee co-chair and applied physics and physics professor Sohrab Ismail-Beigi said. The new policy is designed to give FAS work groups fair access to Yale’s computing resources, he added.

The new rules ask FAS research groups to rank their tasks by priority, Ismail-Beigi said, and redefine the 5 percent share of supercomputer access that faculty research groups have traditionally been granted to account for the different processing capacities that the computer clusters have.

Under the University’s former policy, the majority of FAS research groups were granted 5 percent shares of total access to Yale’s supercomputers, high performance computing specialist Andrew Sherman said. But that figure did not account for differences between the power and speed of various computers, he added.

Sherman explained that smaller FAS work group tasks should be performed on Yale’s smaller supercomputers, while larger tasks requiring more memory should be reserved for Yale’s most powerful supercomputer, Bulldog O. Ismail-Beigi said ITS is working to design a queuing system that will determine which tasks should be conducted on certain machines, and at what times, to maximize efficient use of those computers.

“Heavy hogs should run at low priority and save up some of their time,” Ismail-Beigi said. “We want users to think a bit about what it is they are trying to accomplish.”

Though the new policy has been approved, Sherman said research groups will continue operating under the old system until ITS finalizes its new method for calculating access to supercomputers.

The updated system will weigh the priority levels that FAS research groups assign to their jobs, memory and power requirements, and past resource usage by groups when determining where jobs will fall in the queue, Ismail-Beigi said. For example, he said, a group that constantly files its tasks as high priority would likely use up its 5 percent allocation more quickly than a group that has tasks of different priority levels.

Deputy Provost for Science and Technology Steven Girvin said in an email Monday that the University’s investment in high performance computing has both helped faculty recruitment and advanced faculty research. He said researchers can choose to purchase additional access to supercomputers if they feel a job would require more space than the standard 5 percent, adding that grant funding is available for that purpose.

Sherman said Yale’s supercomputer capabilities have doubled since the University acquired Bulldog O in spring 2011 and made it available for use that fall.

“[Bulldog O] was a big shot in the arm for us,” Sherman said. “We have seen a significant number of research publications and results that have come from using these clusters. [Yale has] hired faculty who insisted on better access to high performance computers.”

More than 100 research groups that use Yale’s five computing clusters, of which 39 groups have a total of 214 individual accounts on Bulldog O alone, Sherman said. Programs conducting research on the machines range from the Physics Department to the School of Music, he said. Sherman added that the Geology & Geophysics research team is granted roughly one-third of Bulldog O’s usage, because the team paid for about one third of the computer’s costs.

Bulldog O can process at 52.53 trillion calculations per second.

Comments