Meet 'Ocelote,' the Fastest Cat on Campus
A new beast inhabits the basement of the Research Data Center: "Ocelote," the latest addition to the UA's central high-performance computing facility.
It doesn't hiss or growl, but it sure looks sleek. Sporting a larger-than-life photo of an ocelot prowling the undergrowth, the front panels cover six racks filled with blinking lights, patch cables and whirring fans.
Borrowing the Spanish word for ocelot, the small wild cat that roams northern and central Mexico, the new supercomputer more than doubles the processing power that the RDC makes available to UA faculty, students and postdoctoral fellows.
Its 336 nodes, each with two 28-core processing units and 192 gigabytes of working memory, are ready to pounce on any highly complex and data-intensive tasks that scientists want to throw at it. The system is highly configurable for custom research needs, including a large memory node with 2 terabytes of memory for crunching especially tough chunks of data.
Ocelote was purchased with funds from the Office of the Chief Information Officer and the Office for Research and Discovery, plus contributions by individual research groups on campus. The system directly aligns with the Never Settle goal of increasing the UA's research capabilities.
"Any researcher on campus can book time on these systems at no charge," said Kelly South, assistant director for communications and marketing in the Office of the CIO. "The Office of the CIO and the Office for Research and Discovery encourage the campus research community to take advantage of all the high-performance computing resources available to them."
"High-performance computing is ideal when you can break your scientific problem up into lots of smaller parts," said Mike Bruck, assistant director for research computing with University Information Technology Systems. One typical scenario, he said, would be a system like Ocelote making the difference in a graduate thesis computational project by completing it in days, rather than months.
"One graduate student in economics needed to go through 500 iterations, each of which would have kept her laptop busy for 24 hours. Instead, she was able to run all the calculations in one night on the UA HPC systems."
The key to the power offered by systems like Ocelote is that its multiple processors work in tandem while also interacting with each other.
Steven Schwartz, a professor in the Department of Chemistry and Biochemistry with appointments in applied mathematics and the BIO5 Institute, said his research would simply not be possible without machines like Ocelote. His group, which contributed significant funding from the Arizona Technology Research Initiative Fund and the National Institutes of Health toward the purchase of Ocelote, develops methods to study complex systems such as enzymes and proteins that make up our heart cells.
"We study in microscopic detail how enzymes in living things work, or how mutations affect proteins that make up cardiac muscle," Schwartz explained. "To do that, we simulate their chemistry, which requires quantum mechanical calculations of hundreds of thousands of atoms at the same time – and that is a big computational tasks. In our cardiac muscle work, we look at millions of atoms; that's a lot of atoms that have to move according to Newton's laws of motion. At the end of the day, none of this happens without a computer."
UITS, in addition to managing and operating the RDC, offers free consulting services to help users get the most out of the high-performance computing resources. Users have the option to purchase their own nodes, too, which grants them extra usage time and high priority access. Also, researchers have the option to install their own servers in the RDC co-location facility.
According to Bruck, the most challenging part, especially for new users, is that these systems operate in Linux command line, while many users are used to operating computers through mouse clicks.
"The system runs in a batch model," Bruck explained. "Users create a script for their particular data and a scheduling software puts those in a queue. The scheduler then plays a big game of Tetris – it fits the incoming jobs with what it is being processed and distributes the tasks across the infrastructure."