two
Volume 5, Number 11 -- March 19, 2008

Intel and Microsoft Look to Jump Start Parallel Computing

Published: March 19, 2008

by Alex Woodie

Microsoft and Intel yesterday announced they will spend millions of dollars to create university research centers at UC Berkeley and the University of Illinois at Urbana-Champaign to jump start parallel computing on the X86-X64 processing platform. If the programs are successful, computer users will benefit from richer digital media, more powerful statistical analyses, and faster mobile applications that can take advantage of the huge processing boosts delivered with multicore processors.

Developing parallel applications that can take full advantage of today's multicore X64 processors is not an easy thing to do. For decades, programmers could count on increasing clock speeds, as defined by Moore's Law, to make their programs go faster without making fundamental changes in their programming techniques.

However, in recent years, as Intel and others reached the thermal limits of conventional chip design, X86 and X64 chip developers were forced to stop ramping up the Gigahertz, and instead resorted to placing multiple processors, or cores, on single chips, or "sockets," to keep up with the demand for ever-faster processors.

Unfortunately, as processor design evolved, there hasn't been a similar shift in software design.Most desktop software applications--and many server applications, too--for X86- and X64-based computers are single-threaded, which means they can't take advantage of today's multicore processors. Considering that most of today's desktops and notebooks sport two or four processors, and that Intel has an 80-core system running in its labs, one could argue that the computing community is woefully behind the curve.

(It's worth noting that servers, operating systems, and applications based on RISC processors have been a bit better at supporting multicore architectures and parallelism. However, with the exception of IBM's POWER architecture, the vast majority of the IT market is focused on so-called "standards-based" computers running X64 servers and Windows or Linux operating systems.)

To keep the mainstream development community from falling even farther behind as multicore computing expands over the next decade, Microsoft and Intel have committed $20 million over the next five years to create two Universal Parallel Computing Research Centers (UPCRCs) at UC Berkeley and the University of Illinois at Urbana-Champaign (UIUC). UIUC will chip in an additional $8 million, while UC Berkeley hopes to get $7 million in state matching funds for industry grants.

Research conducted at the new UPCRCs will focus on advancing parallel programming applications, architecture, and operating systems software, the groups said. Successful research at the UPCRCs will pave the way to allowing programmers to write parallel applications that can take advantage of very fast hardware, without requiring the programmers to become experts in parallel programming.

"We now face the exciting challenge of making parallelism so easy to use that parallel programming becomes synonymous with programming," said Marc Snir, a professor of computer science at UIUC, who will head the new research lab with Wen-Mei Hwu, a professor of electrical and computer engineering.

And if the research is a dud? We may be stuck in a performance backwater, relegated to using archaic, single-threaded applications that don't run any faster on quad-core machines than the single-core process machines of yesteryear.

The universities are bringing together experts on different aspects of IT, including programming languages, compilers, computer architecture, applications, and tools. The UPCRC at UC Berkeley, which will be directed by computer science professor David Patterson, will include 14 faculty members and 50 doctoral students and postdoctoral researchers. The UPCRC will include 20 faculty members (in addition to Snir and Hwu) and 26 graduate students and researchers. "The fruit of that multidisciplinary mixing, plus the tremendous creativity at those places is going to be very powerful for creating those breakthroughs," said Andrew Chien, Intel's vice president of corporate technology group and a director with Intel Research.

Patterson, who made his name for breakthroughs in computer architecture, further elaborated on his plans for the UC Berkeley UPCRC. "To get the real performance we want, we're focusing technology on the 10 percent of the programmers which we call the 'ninja programmers,' the people who are down there in the bits and can really get the most performance and the most out of the power of the chip and deliver it," Patterson says. "So we're going to split our focus, having some for the ninja programmers, and programming assistance for the rest of us, so we can get these exciting apps."

The UPCRC in Illinois will be more mainstream. "We're focusing much more on the programming masses, rather than the ninja," Snir added.

Microsoft and Intel will have special rights to license any patented research generated by the UPCRC programs, according to Tony Hey, corporate vice president of external research at Microsoft Research. "We'll have nonexclusive royalty free access to patents filed on research supported by the center, and also the right to negotiate exclusive license if deemed appropriate," he says. "Otherwise, software will be made freely available under open source license and made available to the whole of IT industry and academia."

UC Berkeley's Patterson doesn't foresee many patents coming from his program. "It's too easy to get around an IT patent," he says. If it wasn't, Stanford University would have raked in billions of dollars for the research it conducted in search technology, which was commercialized by Google, he said. In his 30 years at Berkeley, Patterson says he's never filed a software patent. Indeed, from his comments during yesterday's press conference, Patterson seemed much more willing to work with ninjas than lawyers.

Instead of patents, Patterson sees much of the research being freely distributed to the world through papers, which Microsoft and Intel have placed little to no restrictions on. Patterson also plans to distribute software developed at the UC Berkeley program through the Berkeley Software Distribution (BSD) open source license.




Copyright © 1996-2008 Guild Companies, Inc. All Rights Reserved.
Guild Companies, Inc., 50 Park Terrace East, Suite 8F, New York, NY 10034

Privacy Statement