Skip to main content

Doctoral Student Wins International Competition for Work on Jaguar

By July 26, 2012January 24th, 2013People4 min read

Yuan Tian receives first place in prestigious ACM Student Research Competition for Smart I/O software

The Association for Computing Machinery’s Student Research Competition annually honors student projects that push forward the front line of computer technology. Oak Ridge National Laboratory (ORNL) doctoral student Yuan Tian works to improve user efficiency on one of the world’s fastest supercomputers, so it is no wonder she got first place in the competition.

Yuan Tian, ACM competition winner and ORNL postdoc

“I think one of the major reasons that I won this award is that this project solves a real problem faced by scientists—how to extract knowledge efficiently for data post-processing,” said Tian, who in collaboration with the Oak Ridge Leadership Computing Facility’s (OLCF’s) Scott Klasky and her adviser at Auburn University, Weikaun Yu, developed a tool to help researchers get their data in and out of a supercomputer more efficiently.

One of the largest bottlenecks in supercomputing is the need to transfer massive amounts of data in or out of a system. Klasky leads a team working to streamline this process and was more than happy to work with Tian.

“I was introduced to her through her adviser at Auburn, who is an ex-ORNL employee, and I thought she seemed very enthusiastic,” he said.

To help solve these bottlenecks, Klasky develops input/output (I/O) technology for the OLCF’s Jaguar supercomputer, a Cray XT5 capable of 3.3 quadrillion calculations per second. His work focuses on categorizing, or “chunking,” similar data sets together for quicker access and better organization. Researchers often want to access only a slice of data rather than all of their information, and Klasky’s Adaptable I/O Software (ADIOS) optimizes that process.

Tian began developing her Smart I/O framework within Klasky’s ADIOS software suite. ADIOS provides a simple, flexible way for scientists to describe data produced by their applications that may need to be written, read, or processed outside of the running simulation—saving valuable computing hours. Tian’s Smart I/O enriches this suite of tools.

“Smart I/O is built within the ADIOS framework,” Tian said. “It is able to leverage the existing technologies provided by ADIOS while providing a new method for highly optimized read performance for large-scale scientific data visualization and analysis.”

Tian’s advisor, Yu, introduced her to high-performance computing, and she saw a pressing need to develop more-efficient I/O methods as she continued her studies.

“In general, my motivation to develop high-performance I/O methods is to improve scientific productivity,” Tian said. “Today in HPC we have machines that are getting faster and faster. We have scientific data that are also increasing dramatically in size. How do we store the data efficiently? After the data is stored, how do we extract knowledge from such massive amounts of data? Those are two important questions to be answered by I/O researchers to enable a high-performance end-to-end workflow.”

Without the guidance of supercomputer experts and access to one of America’s fastest supercomputers, Tian felt she may not have been able to win such a prestigious award. “My research has benefited greatly through the collaboration arranged by my advisor, Dr. Weikuan Yu, and the ADIOS team led by Dr. Scott Klasky,” Tian said. “Dr. Yu guided my research step by step and offered me invaluable feedback. As my mentor at ORNL, Dr. Klasky guided me to focus my ideas through his knowledge of science. Without the generous help from ADIOS team members Norbert Podhorszki, Hasan Abbasi, Roselyn Tchoua, Gary Liu, and Jeremy Logan, this project wouldn’t be as successful as it is.”

Tian is graduating with her PhD this summer and will begin to work as a post-doctoral fellow with ORNL’s Remote Data Analysis and Visualization team led by Sean Ahern.—by Eric Gedenk