The Implications of Multicore Processor for High Performance Computing
Michael Perrone · 2008
With some of the largest supercomputing clusters on the planet, the seismic imaging industry has a tremendous appetite for computing resources. Increasing demand for energy and societal pressures for greener energy will only accelerate this growing need. It is therefore clear that seismic imaging companies will be early adopters for the newest generation of supercomputers that enable petascale computing and beyond – machines that are perhaps as close as only months away. However, as this transition to petascale sweeps through the industry, the industry will also be one of the first groups to come face-to-face with the significant changes that will be required in order to go beyond petascale. The industry will have to actively develop plans for how their business models and standard operations will change in order for practitioners have to achieve the performance they require for their evergrowing technical challenges. The next few years will present very important challenges and opportunities for high performance computing in research, industry and business. Petascale computing and, eventually, “exascale” computing will bring the promise of capability to deliver full solutions to some of the most challenging and complex issues facing the industry. However, for well documented technology reasons, these new computing systems architectures will be radically different in design from traditional high performance computing platforms. For example, in response to growing technological obstacles, the processor industry is moving down the multicore path. This development is driving a sea change in the computer industry for which a new "Moore's Law" may be arising - dictating a doubling of the number of cores per unit time. As more cores are squeezed on to a chip, the old programming approaches will not be adequate to achieve the performance required by this industry; and naive assumptions of linear scaling of performance with the number of cores will be very wrong. Recent experience with multicore has identified key challenges which will have to be overcome in order to realize the potential of the next generation of supercomputing. These challenges include fundamental algorithm design; integration of novel architectures with more traditional computational systems; management of the unprecedented amounts of data which are now a key component in all high performance computing activities; and the development, improvement and validation of new applications solutions which address the full complexity of the problems which these novel architectures will make tractable. This presentation will discuss how the industry will be impacted by these changes and what practitioners can do to achieve the full potential of multicore-based petascale and exascale supercomputers.