Robert Barker/University Photography

From left, Johannes Gehrke, Keshev Pingali and David Shalloway, associate directors of the Cornell Theory Center, in front of the V3 cluster.

The Cornell Theory Center (CTC) has fired up its newest and fastest high-performance computer (HPC), called the Velocity-3 Cluster, or V3. The new machine consists of 180 Dell Blades -- machines designed to be servers.

"Blades" are servers that mount vertically in a rack to save space. Each blade contains two processors, creating a machine with a total of 360 parallel processors running the Windows Server 2003 operating system.

Early uses of the V3 cluster have included simulations of the formation of defects in thin metal films and modeling of the process by which liquids turn to solids.

"Because our HPC operations have been growing so rapidly, and our Dell machines seem to never quit, we began to run out of space on our machine floor for new cluster deployments," said Robert Constable, dean of the Faculty of Computing and Information Science. "We looked at blade configurations. Dell's cost was in line, and their service to Cornell has been excellent." Dell, Microsoft and Verari Software Systems worked closely with CTC in building the new cluster.

V3 is just one of several clusters operated by CTC, which now has a total of 1,500 processors arranged in a variety of configurations. These include two clusters -- one with 384 processors and another with 128 -- dedicated to computational biology, and one called ADDM (Advanced Digital Materials Module), belonging to the Computational Materials Institute (CMI), with 340 processors. The ADDM cluster, installed at the same time as V3, can be combined with V3 to form Cornell's fastest supercomputer yet.

The old V2 cluster, with 256 processors, was benchmarked at 1.5 teraflops (trillion floating point operations per second), enough to make it one of the top 100 supercomputers in the world, and the only Windows-based cluster in the top 500. The new combined cluster boosts the speed to 2.1 teraflops.

CTC's first operational cluster, consisting of 256 Pentium III processors, was built in October 1999. Originally known as the AC3 Velocity Cluster, it was renamed V1 after other clusters came online. CTC now uses about 1,250 of its processors in operational clusters. The rest will be used for experimentation, according to Acting Director Tony Ingraffea.

One, Vplus, is currently being used as an experimental Linux cluster, in preparation for the conversion of Velocity 2 to Linux in early 2006. The goal is to make an HPC facility available to researchers who already have demanding applications written for Linux. CTC also is experimenting with clusters of Solaris workstations, and the Agricultural Research Service (ARS) of the U.S. Department of Agriculture is funding the creation of a "small" (36 processors) cluster of Apple G5 Xserve servers, to be hosted by CTC and used by the USDA-ARS US Plant, Soil and Nutrition Laboratory at Cornell for research on gene regulation in bacterial plant pathogens.