NCSA NEWS

To exploit new architectures, first experience new architectures

Story posted October 27, 2006

NCSA has always been fortunate to have an intimate connection to what are among the most successful and ingenious computer science and electrical and computer engineering departments in the country. The University of Illinois at Urbana-Champaign's ECE department has a research budget of more than $25 million per year and has spawned three Nobel winners, the light-emitting diode, and the integrated circuit. The CS program, meanwhile, has 16 active NSF CAREER award winners and 20 IEEE fellows. It also gave the 2000 Turing Award winner and the Chief Technical Officers of Microsoft and Cisco their starts.

It's a rich tradition that we're proud to be a part of.

With the birth of NCSA's Innovative Systems Laboratory last year, we've begun bringing ourselves closer than ever to these pillars of campus. Take the Cell processors recently released by IBM, Sony, and Toshiba. Cell technology was developed with a broad range of applications in mind -- from cryptography to gaming. The Innovative Systems Lab is working closely with Marc Snir, head of the University of Illinois' CS department, and his team. We're getting a few key scientific applications up and running on the system. We're also collaborating with the CS team on trying out the tools they're creating -- like compilers and debuggers -- for the new architecture. Working with these applications and tools gives us the rules of thumb we need to thoroughly describe and improve the performance of a broader range of software in the future.

Field-programmable gate arrays (FPGAs), another area of interest, have been around a bit longer and offer us the opportunity to work with one of the leading lights of the University of Illinois' ECE department. Rather than being hard-wired in predetermined configurations, connections between the basic logical blocks on these specialized chips can be set and reset.

Currently it takes our team at the Innovative Systems Lab several months to get a code up and running efficiently on our existing, experimental FPGA-based systems. That's a large investment, and it typically yields a relatively small improvement in performance. We're working with the ECE department's Wen-Mei Hwu and his team. They're experts in compiler technologies interested in building compilers that will make it easier to port code to FPGAs and to make it run better once it's been ported. Together, we're exploring ways to make applications work across multiple FPGA nodes and ways to measure the performance of those applications.

Admittedly, it will be a while before sweeping scientific findings flow from these collaborations. But it is crucial that we invest in these collaborations now, while we're working with a relatively small number of applications and before we're attempting to scale these technologies up and inject them into production-quality systems.

To exploit new architectures, we have to first experience the new architectures.

In this world of emerging techniques and hardware, NCSA brings technical expertise to the table. But we also bring an interface with application scientists who are engaged in work that could require this type of computing. Both of those things are indispensable. There's a continuum from building new hardware, as companies engaged in building FPGAs or Cell processors do, to building new software, as CS and ECE researchers at UIUC do, to putting applications on those systems and producing new findings, as our users do. NCSA and the Innovative Systems Lab serve as a bridge between these steps, which are too often considered distinct.

That bridge function is a key part of NCSA's march toward petascale applications and systems -- capable of quadrillions of calculations per second. We don't know how to build an effective, efficient, and reliable version of that machine yet. Nor do we know how to exploit a machine like that for science, engineering, and other research endeavors. What we do know is that it all starts here. Collaborations with visionaries and experts like these, whether in building a petascale computer or in applying that power to emerging scientific challenges, are crucial to our success.

Rob Pennington
Chief Technology Officer,
NCSA