SC 2009: The International Conference for High Performance Computing
By Trent Geerdes , ACUS HPC Systems Administrator
The annual Supercomputing conference was held in Portland, Oregon on November 14-20 this year and I was fortunate enough to attend. An estimated 11,000 people from 70 countries attended the The International Conference for High Performance Computing, Networking, Storage, and Analysis. It's an eye opening experience seeing the level of interest and investment internationally in high performance computing related technology. The 131,650 sq. ft. exhibit space was almost overwhelming and easy to get lost in. Vendors and researchers from all corners of the globe were presenting their products, research, and technology. Major industry announcements were made such as Jaguar, the Cray XT5 at Oak Ridge National Laboratory, being upgraded and upsetting IBM's Roadrunner as the fastest computer in the world at running the high performance Linpack benchmark. Intel announced new 6 core Nehalem EX Xeon processors tuned specifically for HPC use. More announcements came from SGI, Cray, Nvidia, PGI, Mellanox, NextIO, and a seemingly endless list of vendors.
Sessions and Papers
SC 2009was a great opportunity to learn from from some of the brightest people at the root of technological innovation. I attended several highly technical tutorial sessions on hybrid MPI/OpenMP, parallel file systems, and OpenCL. The atmosphere was buzzing despite the dry nature of some of the subject matter due to the expert credentials of the presenters and the open format of the presentations. I attended a couple of sessions related to the Lustre file system that were very useful and was able to interface directly with Sun engineers. A huge benefit of attending SC were the opportunities to talk Lustre with individuals in the room from New Zealand, Australia, England, Canada, and around the US. Most enjoyable and impressive to me was the OpenCL session presented by individuals from Khronos group members AMD, Nvidia, and Intel. Many papers on recent GPU based performance achievements were presented along with energy efficiency and data visualization playing increasing roles at the conference. Clearly GPU acceleration is making big waves in HPC circles as China's GPU-CPU Tianhe-1 hit number five in the Top 500 during the conference. Industry players like PGI also announced GPU compilers. Most of the GPU cluster activity isn't yet happening on production systems but it seems that may change over the next year or two. SC is a unique environment to network with and learn from other people in highly specialized fields and I predict it will continue to grow.
Back at Home
The HPC (High Performance Computing) Team has been hard at work the last few months readying the new 224 node Talon cluster. It is a massive increase in computational resource from what UNT has been able to offer historically. In 2006 Dr. Philip Baczewski wrote about the usage of available HPC resources and in 2005 about 28.5 CPU years were used. The Talon cluster has ample resources that if utilized, could surpass that on a monthly basis. We have been working with several research faculty to iron out some last production environment details and are now days away from being ready to let our talented researchers begin utilizing this resource (see the HPC Ribbon Cutting article in this issue for pictures). We are excited to be helping to achieve new levels of computational research and to further UNT's research.
Questions or comments can be sent to Trent Geerdes, HPC Systems Administrator, CITC ACUS.