News

Purdue Builds 1,200-Node Supercomputer Before Lunch

• By Dian Schaffhauser
• 08/03/09

Purdue University in West Lafayette, IN turned the installation of its latest research supercomputer into a time trial when 200 participants helped build what is expected to be Purdue's largest campus computer. Before lunchtime, 200 staff and volunteers from Purdue and several other universities, technology vendors, and integration partners had unboxed and assembled the components of what has been named the Coates Cluster. The cluster is named after Ben Coates, the former head of the School of Electrical and Computer Engineering at Purdue, who pushed for high performance computing and networking on campus. He died in 2000.

The new Coates Cluster consists of 1,200 HP dual-quad core computer nodes with a total of 9,000 AMD processors, organized into sub-clusters, each with a different memory and storage configuration. All nodes have 10 gigabit Ethernet (10GigE) within a network infrastructure running Cisco Nexus data center switches and Chelsio iWARP Ethernet remote direct memory access adapters. The nodes will run Red Hat Enterprise Linux 5 and use Altair Engineering's PBSPro 10.x for resource and job management.

"Last year we unboxed the components for our Steele supercomputer in the morning and we were doing science in the afternoon," said Gerry McCartney, CIO and vice president for IT. "We expect to do the same thing with Coates, even though it is significantly larger." The Steele supercomputer, also assembled in a day, has 6,500 cores.

To generate interest in the latest build-out, Purdue's IT staff created a parody movie trailer, "Cores," a take-off on Pixar's "Cars."

Coates is expected to rank in the top 50 supercomputers worldwide as determined by TOP500. It's expected to have a peak performance of 90 teraflops.

Coates, like Steele, was built as a "community cluster," in which faculty on campus contributed research funds to fund the purchase, said John Campbell, associate vice president for Rosen Center for Advanced Computing at Purdue, where the supercomputers are located. "Besides the cost savings from making a group purchase, the faculty can borrow computing cycles from other faculty when the other clusters are idle. This gives the researchers more flexibility, and we also have unused computing cycles we can offer to the National Science Foundation's TeraGrid."