Perspectives on HPC and the SC Series of Conferences
Once a year, the leading experts from the world of high performance computing gather at SC to assess the current state of HPC and to look ahead to what the future holds. These are the people creating the technologies that will shape our lives. This year, as the conference celebrates an amazing 20 years, several industry thought leaders and long-time attendees reflect on what is most important to them.
Dan Reed, Director of Scalable and Multicore Computing Strategy at Microsoft
The SC conference continues to grow in scale, scope and variety, with a diverse set of workshops, plenary speakers, technical program sessions and, of course, the massive exhibit floor. In addition to the public program, there are a seemingly endless series of sidebar meetings and lots of technical socializing. Take advantage of the fact that you can talk to almost anyone connected to highperformance computing during the conference, but remember that you can overdo it and never be seen at any of the official venues!
Undoubtedly, one of the great hallway discussion topics will be the effect of the economic downturn on HPC research, infrastructure acquisitions and vendor finances. It is quite possible that some startups and smaller companies may not survive. For those in the U.S., the Presidential transition and the implications for research funding will also be hot topics.
Finally, I suspect two other discussions will center on the relationship between academic Grids and commercial clouds and the relationship between trans-petascale (exascale) options and the design of extremely large data centers. The latter is deeply connected to ecofriendly computing system design and energy efficiency. Answers these questions will affect the future of large-scale computing, our research investments, user communities and the types of applications we can support efficiently.
Remember — bring your running shoes. Your feet will thank you later.
Pete Ungaro, President and CEO of Cray Inc.
Without a doubt, SC is the most important conference of the year for our community. This year marks the 20th anniversary of the conference but even more interesting is that this year’s event will officially kick off the start of a new era in HPC — the petascale era.
There will be a lot going on at the conference around petascale computing — vendors and customers highlighting their capabilities as well as end-users contemplating what will now be possible with this new-found power. A related and important theme will be green computing, especially how can petascale systems be built in a way that minimizes their impact on the environment. Cray, of course, is no exception — we are very excited about bringing petascale computing to our customers through our scalable system designs and innovative power and cooling technologies. We believe that the petascale era promises to enable significant technological breakthroughs as scientists and engineers are able to tackle larger problems with higher fidelity.
Be sure to take a few minutes to stop by our booth to see how we’re tackling the petascale challenge as well as bringing Cray supercomputing technology to individual users. Have a great conference!
Debra Goldfarb, President and CEO of Tabor Communications
Being an industry observer, I have seen a lot of change. Undoubtedly, we are entering an exciting innovation cycle in terms of technology, usage models and access. This year I have a few “rules of the road” which will guide my week in Austin:
Spend time on the periphery. there is a lot of interesting stuff to see which is not in the main hall, but rather in the small booths which sit out on the edges. This is where you can often get a window into “what’s next.” I will be looking for technology which enables productivity such as: appliances (application as well as infrastructure); development tools; application frameworks; energy efficiency concepts; and more adaptive access models (such as cloud or other webservice models).
Explore “Edge” HPC. Tabor Research is researching the use of HPC technologies and concepts outside of science and engineering. These include virtual worlds, ultra-scale infrastructure (such as search), complex event processing, and business optimization (such as real-time data mining). My goal is to better understand requirements, application evolution, and most importantly, what is in the “envelope” and what falls out.
The politics of science. Timing is everything and given the recent (and quite extraordinary) change in administration, it will be fascinating to get a read on what this means to this community. And, by the way, it should mean a lot in terms of priorities — in science, technology, industry and education.
Jack Dongarra, Distinguished Professor of EECS at the University of Tennessee
I have attended all of the SC meetings, and won’t miss it for the world; it represents “Homecoming Week” for High Performance Computing.
This is truly an awesome time for high performance computing and computational science research, with a number of systems achieved performance exceeding the PFlop/s mark. There are a number of interesting problems that will need to be overcome as we are faced with systems with greater than a million threads of execution. Advancing to the next stage of growth for computational simulation and modeling will require us to solve basic research problems in Computer Science and Applied Mathematics at the same time as we create and promulgate a new paradigm for the development of scientific software.
To make progress on both fronts simultaneously will require a level of sustained, interdisciplinary collaboration among the core research communities that, in the past, has only been achieved by forming and supporting research centers dedicated to such a common purpose.
I see five important areas that will need attention; effective use of manycore and hybrid architectures, exploiting mixed precision in the algorithms, self adapting and auto tuning of software, fault tolerant algorithms, and communication avoiding algorithms.
William Feiereisen, Director DoD High Performance Computing, Lockheed Martin
I have attended most of the Supercomputing conferences since the 1980s. They always been one of the central opportunities to gather with virtually the entire scientific computing community and to see the latest developments in the spectrum of technologies that interact to make the field of supercomputing.
All of this is available in one place each year, everything from the applications and the important problems that they solve to the latest hardware upon which they run. There is a flavor of computational sciences which has always been my motivation and excitement about the field, but I also confess to not being immune to the latest raw hardware speed breakthroughs presented by each of the manufacturers.
I always plan my week around three things: the technical sessions and tutorials; the exhibits on the show floor; and increasingly in recent years, everyone else who attends and the possibility of much personal interaction. When I started my career, the technical sessions and tutorials dominated my time at SC, however I find much rewarding time is now spent in conversation over convention center coffee. Over the years I believe that many connections and ideas have been hatched at SC in just this way. There is a critical mass that gathers here each year and supports this atmosphere.
It’s for these reasons that I keep coming back each year and I look forward again this year to spending the week in Austin.
Marc Snir, Co-director, Universal Parallel Computing Research Center, University of Illinois
Moore’s law does not mean, anymore, ever increasing processor performance; instead, it now means an ever increasing number of processors on a chip. Just waiting for processor performance to catch up to your needs is not an option, anymore; the only way to increase application performance, is to parallelize the application and scale it to an increasing number of processors.
This is a major new challenge. On the positive side, parallel programming is moving from being an esoteric art practiced by few experts into a a mainstream occupation. It becomes a major concern of large companies, such as Microsoft and Intel (see, for example, their investment in the Universal Parallel Computing Research Centers at Illinois and Berkeley).
This is an opportunity for the HPC community: Rather than building support for parallelism on top of sequential languages and programming environments, it becomes now possible to scaleup languages and environments that are build up-front to support parallelism and that are supported by massive investments.