Virginia Tech will present solutions to two major challenges to high-performance computing--efficient storage and energy use--at the Supercomputing 07 (SC|07) Conference Nov. 12-15 at the Reno-Sparks Convention Center.

And Virginia Tech will use the event to announce the Green500 List, which ranks the most energy-efficient supercomputers in the world. The university will also have presentations on visualization, network interfaces, and thermal self-awareness.

SC|07, sponsored by the Association for Computing Machinery (ACM) and the IEEE Computer Society, will showcase how high-performance computing, networking, storage and analysis lead to advances in research, education and commerce.

A Virginia Tech team is among the finalists presenting solutions to an important high-performance computing problem. Limited bandwidth and capacity for data storage can limit what supercomputers and grids can achieve. There has been a storage challenge at the last two supercomputing conferences. The competition showcases effective approaches using the storage subsystem with actual applications. Participants must describe their implementations and present measurements of performance, scalability, and storage subsystem use.

Wu-chun Feng, associate professor of computer science and electrical and computer engineering at Virginia Tech; Pavan Balaji, post-doctoral researcher in the Mathematics and Computer Science Division at the Argonne National Laboratory; and Jeremy Archuleta, Ph.D. student in computer science at Virginia Tech, teamed up to create ParaMEDIC: A Parallel Meta-data Environment for Distributed I/O and Computing.

ParaMEDIC is an environment that decouples computation and input/output (I/O) in applications and drastically reduces I/O overhead through metadata processing. They have demonstrated a five-fold improvement on the Teragrid and a 25 fold improvement between Argonne National Laboratory and Virginia Tech when applying ParaMEDIC to mpiBLAST, an open-source parallelization of the Pairwise BLAST DNA sequence-search library.

The team will present its solution at 1:30 p.m. on Tuesday, Nov. 13, in rooms A10 / A11 at the convention center. There will also be a multi-monitor visualization of the mpiBLAST data from this finalist entry at Virginia Tech’s booth.

A new challenge to large datacenters being addressed by Virginia Tech researchers is energy consumption. Supercomputers consume as much power as a small city. Kirk Cameron, associate professor of computer science at Virginia Tech, who pioneered the area of high-performance, power-aware computing, is chairing a session on the energy crisis. Panelists from the Environmental Protection Agency, IBM, the Tokyo Institute of Technology, Oak Ridge National Laboratory, and Lawrence Berkeley National Laboratory will present their thoughts and ideas regarding the future impact of power on the SC community from political policy such as the EPA Energy Star program to facility requirements and systems design. The program is 8:30 to 10 a.m. Friday, Nov. 16, in rooms A1 / A6.

Virginia Tech computer science Ph.D. student Rong Ge will present her research on Improving Power-Performance Efficiency in High-End Computing at the Doctoral Research Showcase, at 3:30 p.m. Wednesday, Nov. 14, in room A10 / A11. Ge is a member of the Center for High-End Computing Systems at Virginia Tech. She reports that the center’s theoretical models can improve algorithm performance by 59 percent; while her hardware/software toolset for power profiling provides previously unavailable insight to parallel scientific application power consumption; and power-aware techniques can save 36 percent energy with little performance degradation.

Feng and Balaji are part of yet another team that is addressing high-performance network interfaces. Due to the growing need to tolerate network faults and congestion in high-end computing systems, supporting multiple network communication paths is becoming increasingly important. However, multi-path communication comes with the disadvantage of out-of-order arrival of packets (because packets may traverse different paths). Pavan Balaji, Wu-chun Feng, Sitha Bhagvat of Dell Inc., Dhabaleswar Panda of Ohio State University, and Rajeev Thakur and William Gropp, both of Argonne National Laboratory, will analyze the trade-offs in designing a feature-complete Internet Wide-Area RDMA Protocol (iWARP) stack that provides support for out-of-order arriving packets and multi-path systems, while focusing on the performance of in-order communication. The presentation is at 3:30 p.m. Wednesday, Nov. 14, in A2 / A5.

Three Virginia Tech computer scientists will present a poster on Thermal-aware High Performance Computing. Graduate students Hari Pyla and Dong Li and Associate Professor Cameron created an infrastructure to gather data from temperature sensors, correlate the data to source code, and control the thermal characteristics of an application at runtime. Their results indicate that thermal throttling can be accomplished using a systemic controller. Posters will be displayed Tuesday, Nov. 13, from 5:15 to 7 p.m., in the ballroom lobby.

Kevin Shinpaugh, director of high performance computing at Virginia Tech, organized presentations for the university’s exhibit booth (2803) that showcase scientific computing at Virginia Tech, including demos on visualization by Nicholas Polys, director of visualization for research computing, and a multi-monitor display of the mpiBLAST data from the entry in the storage competition. The booth is sponsored by the Offices of the Vice Presidents for IT and Research at Virginia Tech.

Virginia Tech is a comprehensive research university with the largest full-time student population in Virginia. Virginia Tech Advanced Research Computing hosts several high performance computer systems, including System X and three SGI Altix Clusters. The university is a founding National LambdaRail (NLR) Class A member with responsibility for facilitating all access to the NLR Washington DC node for Virginia, Maryland, and Washington, D.C.

Contact:

Share this story