The transformational impact of supercomputing in science and engineering has been remarkable, especially in simulation-based science. As supercomputing systems have reached tera-scale and now peta-scale, these orders-of-magnitude increases in computational capabilities have enabled higher resolution simulations, more exploration of wide parameter spaces, and even multi-scale, multi-physics simulations—all leading to more realism, and more understanding. However, many important problems require yet more orders of capability, and are driving new technologies such as GPUs, MICs, FPGAs, and advances in storage and networking. The supercomputing community is again rich in component technologies being used and explored, with more than a dozen processor technologies alone now being used or developed for supercomputing systems.

Concurrently, the explosion in digital data creation—from computing systems and networks, but also from new generations of digital detectors in all domains—is driving a corresponding explosion in data-driven science. Bio-informatics, environmental science, health care, and social sciences are relatively new communities using massive computing systems, with new usage models and requirements influencing the design, configuration, and operation of large-scale systems.

In this talk, we will examine the emerging technology trends that are changing the face of supercomputing, and discuss expanding role of supercomputing systems in a world dealing with ‘big data’ as well as ‘big simulations.’ We will thus explore how supercomputing is not becoming less relevant but even more so due to ‘big data,’ and how its very definition, as well as scope and impact, will grow in the years ahead.

Compute performance increased by orders of magnitude in the last few decades, made possible by continued technology scaling, increasing frequency, providing integration capacity to realize novel architectures, and reducing energy to keep power dissipation within limit. The technology treadmill will continue, and one would expect to reach Exa-scale level performance this decade; however, it’s the same Physics that helped you in the past will now pose some barriers—Business as usual will not be an option. The energy and power will pose as a major challenge—an Exa-scale machine would consume in excess of a Giga-watt! Memory & communication bandwidth with conventional technology would be prohibitive. Orders of magnitude increased parallelism, let alone explosion of parallelism created by energy saving techniques, would increase unreliability. And programming system will be posed with even severe challenge of harnessing the performance with concurrency. This talk will discuss potential solutions in all disciplines, such as circuit design, test, architecture, system design, programming system, and resiliency to pave the road towards Exa-scale performance

Future systems in HPC and cloud computing will likely lead to billion way parallelism, placing unprecedented demands on data handling. In this talk we will quickly review the existing paradigms in managing data and point out some of the challenges associated with scaling these up.   We will discuss that a deeper integration between applications and data, discovered by dozens of scientists all around the world, all the way through the storage stack from CPU caches to disk, can eliminate many of these obstacles.  The emerging architecture is quite different from current practice, but appears to have a very manageable API.  Key examples involve data placement and data access patterns, failure handling, peer-to-peer decision making on collective operations, and information life cycle management with guided mechanisms.  The implementation to fully adopt this architecture would lead to changes all the way from the compiler level, to storage behaviors that haven't been challenged since the emergence of Unix.