Bioengineering professor David Gough (left), principal investigator of NBCR Peter Arzberger, Bioengineering vice chair Andrew McCulloch, director of SDSC's Advanced CyberInfrastructure Lab Phil Papadopoulos, Dell account manager Doug Shaw, and Bioengineering chair Shu Chien perform a ceremonial ribbon cutting at the dedication of UCSD's new cluster computer.
"No single computational model spans all these biological scales, but this powerful new cluster will enable us to integrate models over many of these scales, which will make it possible for us to predict, in some cases, the clinical consequences of specific genetic mutations or biochemical alterations caused by disease," said Andrew McCulloch, a professor and vice chair of the Jacobs School's Department of Bioengineering. He celebrated the installation of the cluster at a ribbon-cutting ceremony March 9 in the basement of Powell-Focht Bioengineering Hall with fellow project co-leader Peter Arzberger, principal investigator and director of the National Biomedical Computational Resource (NBCR), a program funded by the National Institutes of Health, and director of Life Sciences Initiatives at UCSD.
Arzberger said the Department of Bioengineering and NBCR Dell Rocks Cluster will be quickly integrated into a computational grid to provide the resource to as many UCSD researchers as possible. Eventually, the cluster will also be made available to computational biologists and bioengineers across the country as part of a new paradigm often referred to as grid computing.
The new cluster, which was purchased from Dell for less than $180,000 per Teraflops (trillion floating point operations per second), is a distributed-memory parallel computer. It is valued not only for its speed, but also because its 428 gigabytes of memory capacity and 20 terabytes of storage will enable researchers to solve ever larger and more sophisticated problems. "It's not always just a matter of computational speed with some of our models -- some are memory limited," said McCulloch. "For example, our models of the propogation of an electrical impulse thorough the heart wall require us to generate matricies with millions of individual cells, and our ability to solve such problems is limited by the memory available. This new cluster will enable completely new simulations of the heart and other biological systems."
Traditional supercomputer centers have historically been the workhorses of computational biology, but McCulloch said the drive to build ever faster and more specialized supercomputers has required computational biologists to continually reinvent their modeling software. "It's not new that science has been developing large computational challenges, but what is new is the commoditization of supercomputing," said McCulloch. "These commodity clusters running with an open source operating system and management software can provide a stunning degree of performance that a few years ago simply couldn't be found in a small room in the basement of a bioengineering building."
The cluster's management tool is called Rocks, an award winning toolkit developed at UCSD by a team led by Philip Papadopoulos, director of the Advanced CyberInfrastructure Lab at the San Diego Supercomputer Center and associate research scientist in the Department of Computer Science and Engineering. The open-source Rocks toolkit has been designed for rapid and scalable deployment of clusters. It currently is used in more than 500 supercomputing clusters worldwide and supports the 64-bit processors in the new cluster. In addition to its many other features, Rocks simplifies many administration requirements.
Within a few days of unpacking more than 400 boxes containing the Dell nodes, bioengineering researchers this week had modeled how electrical impulses trigger contractions in the heart of a pig.