Over at the ISC Blog, Mark Parsons from the EPCC supercomputing centre writes that scalable software is the real Grand Challenge of Exascale.
I believe that the problems that we’ve seen at the Petascale with regard to the scaling of many codes are insurmountable if we take the incremental change approach at the Exascale. Looking at the CRESTA codes, it is highly unlikely any of them will scale to the Exascale, even allowing for weak scaling (through increased resolution of the model under study) using incremental improvements. This means we need to think about disruptive changes to codes in order to meet the challenge.
Parsons leads the CRESTA FP7 project, which is focussing its work on a small set of six HPC applications that are widely used today and represent the sort of codes that will have to run on Exascale systems. He says that over the past 20 years, the community has managed to cope with each new generation of hardware through incrementally improving our codes. But today, simply changing a solver or some other disruptive change to an existing code will not be enough.
We simply do not understand how to compute using one billion parallel threads (except perhaps in trivial cases). It requires us to completely rethink how we simulate our physical world using this much parallelism. The problem goes to the foundations of modern modelling and simulation – we need to think beyond the tools we have today and invent new methods to express the mathematical descriptions of the physical world around us, on these and even larger systems in the future. Only by doing this will we move modelling and simulation forward for the next 20 years. This is the real challenge we face at the Exascale.
Over at The Exascale Report, Mike Bernhardt has posted an interesting interview with Intel’s Raj Hazra, Vice-President Intel Architecture Group & GM Technical Computing. When asked why we should build an Exaflop machine, Hazra contends that we should focus on the economic benefits.
So the exciting part of exaFLOPS is that while it allows you to build those great big machines, it also lets you democratize HPC by essentially building half a petaflops, which today would be in the top 100 supercomputers, under everybody’s desk. And it would be done at both a capital and a TCO cost that changes the game. The economic discussion is fundamental. That’s 300,000 potential users just in manufacturing in the U.S. that we’ve looked at. If they were to use any level of HPC or extreme scale computing, how much value would they generate and what would it do to the HPC market?
In the past years one could observe a significant consolidation of non-system vendors providing interconnect technologies and associated topological options. Only a few providers with proprietary interconnects are still around. The two talks in the session will provide a deeper insight into future interconnect standards and topologies and will also address the migration from electrical to optical interconnects at system level.
Over at The Exascale Report, Mike Bernhardt has posted an in-depth with Intel’s Diane Bryant, Senior Vice President and General Manger, Datacenter and Connected Systems Group. Bryant gives her perspective on high performance computing, U.S. competitiveness, and the goal of reaching Exascale computing performance by 2020.
The U.S. government’s investment in the move to Exascale is very important. The U.S. government has been a driving force year after year after year in advancing the state of computing. That’s important for the industry. It’s important for the United States, for public interest, and for economic development. And, at large, we will collectively, holistically continue the move to exascale. Government can’t do this without industry, and industry can’t do this without the support of the government. Intel invested $10 billion in R&D last year, and we continue to make a significant investments year over year in R&D. We take the High Performance Computing and the supercomputing space extremely seriously. If for no other reason, technology starts at the top and like water, falls down. Everyone benefits from that trickle down – from the highest supercomputing investment down to commercial HPC down to the desktop. So we will continue to invest. From an international perspective, you certainly do see other governments – other geographies – recognizing the criticality of an investment by the government in supercomputing. It’s actually very encouraging. It reinforces what the U.S. government has been doing year after year, decade after decade.
Over at the ScaleOut Blog, Rob Mitchum writes that Argonne’s Rick Stevens made a great case for Exascale at the recent hearing on Capital Hill.
We have identified five major hurdles that must be overcome if we are to achieve our goal of pushing the computing performance frontier to the Exascale by the end of the decade:
We must reduce system power consumption by at least a factor of 50.
We must improve memory performance and lower cost by a factor of 100.
We must improve our ability to program systems with dramatically increased levels of parallelism.
We must increase the parallelism of our applications software, math librareis and operating systems by at least a factor of 1,000.
We must improve systems reliability by at least a factor of 10.
These are not simple tasks. But all of us who are working in this community believe that Exascale supercomputing will be a reality by the end of this decade. It will happen first in the U.S. if we can get the investment needed. This bill is a great start to that commitment. Ultimately, this is a race, not against our international competitors, but rather it’s a race for us. Exascale computing is necessary to the achievement of our most urgent goals in energy, in medicine, in science and in the environment. And it will have a profound impact on industry competitiveness and national security. I believe we have a duty to move as swiftly as we can.
Over at the ISC Blog, Thomas Lippert from the Jülich Supercomputing Centre writes that the DEEP project is about to demonstrate that the pitfalls of Amdahl’s law can be avoided in specific situations.
The applications adapted to DEEP are selected in order to investigate and demonstrate the usefulness of the combination of hardware, system software and the programming model to leave ground and leap beyond the limits of Amdahl’s law of parallel computing. We are eager to show our first results at the ISC’13 in Leipzig.
In this video, the House Committee on Science, Space, and Technology’s Subcommittee on Energy holds a May 22 hearing to examine HPC research and development challenges and opportunities, specifically as they relate to exascale computing.
Testifying before the Subcommittee were Dr. Roscoe Giles, Chairman of the Advanced Scientific Computing Advisory Committee; Dr. Rick Stevens, Associate Laboratory Director for Computing, Environment and Life Sciences at Argonne National Laboratory; Ms. Dona Crawford, Associate Director for Computation at Lawrence Livermore National Laboratory; and Dr. Daniel Reed, Vice President for Research and Economic Development at the University of Iowa.
Exascale computing will be an important part of a larger effort to improve the U.S.’s overall high-end computing capability to address a broad range of academic, industrial, and national security needs. While research in next generation computing architecture and software continues to require strategic government investments, Members also explored the significant economic benefits that can arise from full utilization of existing high performance computing capabilities in ongoing scientific research.
Over at ExtremeTech, Joel Hruska writes that the daunting challenges of achieving exascale compute levels by the end of the decade were brought home recently in a presentation by Horst Simon, the Deputy Director at NERSC. In fact, Simon has wagered $2000 of his own money that we wont get there by 2020.
But here’s the thing: What if the focus on “exascale” is actually the wrong way to look at the problem?
FLOPS has persisted as a metric in supercomputing even as core counts and system density has risen, but the peak performance of a supercomputer may be a poor measure of its usefulness. The ability to efficiently utilize a subset of the system’s total performance capability is extremely important. In the long term, FLOPS are easier than moving data across nodes. Taking advantage of parallelism becomes even more important. Keeping data local is a better way to save power than spreading the workload across nodes, because as node counts rise, concurrency consumes an increasing percentage of total system power.
Japan News reports that the country’s science ministry is considering development of an exascale supercomputer that would be 100 times faster than K computer, which is currently the nation’s fastest machine. With a goal of completing the machine by about 2020, the Education, Culture, Sports, Science and Technology Ministry is preparing to request funding for conceptual designs and other areas in next fiscal year’s budget, the sources said.
Exascale computer projects are already under way in the United States, Europe and China, all aiming for completion around 2020. The working group decided to enter the fierce international race to develop an exascale supercomputer because “it would aid scientific and technological development, and help improve industrial competitiveness,” the sources said.
Electronics Weekly reports that the Barcelona Supercomputing Center is working with Intel to set up a research lab in Spain to develop technologies needed for future exascale supercomputers with up to 100 million processor cores
The BSC Exascale Laboratory will research scalable parallel run-time systems that are needed to support these very high levels of parallel computing.
BSC is one of Europe’s most renowned HPC labs and offers very interesting technology to scale run time systems, tools and applications up to exascale level,” said Stephen Pawlowski, Intel senior fellow.
There is a fierce competition on the storage market to offer the best performing devices, with great management at a low price. The EIOW group, from the outset, decided that it would not attempt to offer an end-to-end solution, which would necessarily involve competing instead of working with storage providers. The focus of EIOW is on middleware to provide, for example, schemas describing data structure and layout, novel access methods to data for applications, a uniform data management infrastructure and a framework for the implementation of layered I/O software, similar in spirit to HDF5 as a specialized use of a parallel file system. We decided EIOW should be open, and have interfaces to layer on lower level storage infrastructure such as object stores, databases and file systems as provided by storage providers, to allow their expertise and leadership in this area to continue to benefit the HPC community.
Back in July 2012, Whamcloud was awarded the Storage and I/O Research & Development subcontract for the Department of Energy’s FastForward program. Shortly afterward, the company was acquired by Intel. The two-year contract scope includes key R&D necessary for a new object storage paradigm for HPC exascale computing, and the developed technology will also address next-generation storage mechanisms required by the Big Data market.
The subcontract incorporates application I/O expertise from the HDF Group, system I/O and I/O aggregation expertise from EMC Corporation, object storage expertise from DDN, and scale testing facilities from Cray, teamed with file system, architecture, and project management skills from Whamcloud. All components developed in the project will be open sourced and benefit the entire Lustre community.
The DOE Exascale Mathematics Working Group (EMWG) has issued a Call for Position Papers. Selected contributors may have the opportunity to participate in the Workshop on Applied Mathematics Research for Exascale Computing, currently planned for August 21-22, 2013 in Washington, D.C.
The ENWG was formed for the purpose of identifying mathematics and algorithms research opportunities that will enable scientific applications to harness the potential of exascale computing.
While opening up the possibility of conducting ground-breaking science, computing at the exascale will introduce difficult challenges such as extreme concurrency, memory and data motion limitations, energy control, and resilience. Substantial applied mathematics research is required to realize the full benefits of computing at the exascale.
This week, Scientific Computing is featuring an interview with Jack Dongarra from the University of Tennessee. Dongarra discusses the origin of the LINPACK benchmark and why multidimensional barriers to Exascale computing will force a disruptive change in the form, function and interoperability of future software infrastructure components.
As a number of recent studies make clear, technology trends over the next decade — broadly speaking, increases of 1000X in capability over today’s most massive computing systems, in multiple dimensions, as well as increases of similar scale in data volumes — will force a disruptive change in the form, function and interoperability of future software infrastructure components (I’ll call this the X-stack) and the system architectures incorporating them.
Dongarra has several talks on Exascale topics coming up at the ISC’13 conference, which takes place June 16-20 in Leipzig, Germany. Read the Full Story.
Over at Computerworld, Patrick Thibodeau writes that President Barack Obama’s federal budget plan released this week calls for increases in federal R&D spending, but does not set a goal for building Exascale systems.
Every decade, compute power has double in alignment with Moore’s Law. The power of supercomputing has followed a similar path. Following that target, an exascale system might have been possible in the 2018 timeframe. The petaflop barrier was broken in 2008. Exascale development, however, poses unique challenges because of the enormous amount power such a system would likely need. That challenge is driving a re-thinking of all aspects of systems development. But because of funding uncertainty, federal officials have said that an exascale system may not be developed in the U.S. until 2022.