Wednesday Keynote: HPC Achievement and Impact – 2009

Chair:

Dr. Frank Baetke, Global HPC Technology Program Manager, Hewlett-Packard, Germany

Keynoter:

Prof. Dr. Thomas Sterling, Arnaud & Edwards Professor of Computer Science, Louisiana State University, USA

It is the beginning of the Petaflops Era. Long awaited and for which many have endeavored, this year just concluded has been the first of a decade long epoch for which sustained performance will be measured in Petaflops.

At least two machines in the US are now Petaflops capable and a number of systems around the world are in various planning stages to deliver this unprecedented scale. Both IBM and Cray are exploiting advances in multicore and heterogeneous architectures to find the sweet spot between applications, sustained performance, power consumption, and cost. In the meantime major new studies in the pathfinding domain of Exaflops computing greatly extends the earliest DARPA study reported last year to the areas of systems software again under the sponsorship of DARPA and in important scientific and technical applications supported by the US DOE. The challenging confronting HPC is the effective programming of multicore architectures and over the last year, significant advances have been achieved that will be discussed.

Similarly, programming methods for heterogeneity such as the IBM cell architecture used in Roadrunner and the NVIDIA GPU architectures have been devised to make better use of these important technologies. New trends in distributed environments have seen growing interest including their relevance to HPC. Clouds make available a diversity of computing, storage, communications, and services resources on demand. But the question of how Clouds may help improve the effectiveness of our field is less clear and still being explored. This etesian presentation will discuss all of these important advances and events of the past year and consider what is in store for HPC in the year to come.

This presentation will review the highlights of the last year in High Performance Computing considering advances in hardware technologies and systems as well as progress in software development and other key accomplishments. This may prove to be the last year before the inauguration of the Petaflops era with at least one machine demonstrating half a Petaflops peak capability and a number of such systems within 10% of that long awaited milestone. New machines from Cray and IBM are under development to surpass that level as well as by NEC. During the last year, Japan has announced its strategy for realizing general purpose Petaflops scale. The US National Science Foundation has awarded its first grant for a Petaflops scale system to NCSA at the University of Illinois while the US Department of Energy prepares to install more than one such machine of different designs at their national laboratories. Even as the world awaits the deployment of real Petaflops computers some 14 years after they were first conceived in 1994, experts have convened in two separate studies to investigate the potential impact and most promising approaches to achieving a thousand times that before the end of the next decade. These exciting events and many others in our rapidly advancing field will be discussed during this presentation.