Opinion: The time for parallel computing is now

The electronics industry is not investing enough money, time and energy to address what has become a pressing need for a genuine technology breakthrough in parallel programming. It is time to fund multiple large-scale projects in this area and stakeholders need to step up to the plate quickly.

David A. Patterson, a professor and researcher at the University of California at Berkeley, outlined the gravity of the problem in a recent talk.

Microprocessor makers have hit a "power wall" that undermines their conventional design techniques and has driven them to multicore architectures. But this just shifts the problem to the need for parallel software, something researchers have pursued with little success for the past 40 years.

If researchers do not solve this problem now, multicore architectures will soon outstrip the ability of software to keep up. If new software cannot take advantage of improved chips, the growth of the computer industry—and the many industries that depend on it—is in peril.

Insufficient effort
The response to this challenge from the industry's large, aggressive competitors has so far been wimpy.

Intel and Microsoft put together a reportedly $10 million, five-year university research grant to tackle the issue. The duo did a good job raising the level of consciousness about the problem by seeking competitive bids from 25 top computer research institutions for the grant. Still, their effort is tiny in comparison to the magnitude of the problem and their resources.

Intel and Microsoft are two of the most profitable companies in the sector, having worked for years to drive most of the value of computers into their processors and Windows. Now it's time they dig deep.

Intel has already funded some efforts to help train the next-generation of programmers in the latest parallel techniques and Microsoft has hired a handful of heavy hitters in parallel programming including Burton Smith and Dan Reed. But William Dally, professor of computer science at Stanford University, was right when he said the industry needs "to start experimenting right away and try a dozen different ideas to find a few that work."

The Intel-Microsoft tie-up should not just tease multiple researchers with a $10 million grant awarded to one institution. They need to significantly up the ante and fund multiple efforts. Ten million is a drop in the bucket of the R&D budgets at Intel and Microsoft. You have to wonder about who is piloting the ship in Redmond these days when the company can afford a $44 billion bid for Yahoo to try to bolster its position in Web search but only spends $10 million to attack a needed breakthrough to save its core Windows business.

Intel and Microsoft are savvy enough to know you cannot schedule a breakthrough. Both companies have put competing internal teams in place on difficult projects in the past.

Either top management at these companies fails to fully appreciate the magnitude of the technology hurdle they face, or they are hoping for breakthrough research on the cheap. Either way, they need to step up or be dragged to the plate.

It's not just Intel and Microsoft that has skin in this game, although these highly profitable companies have plenty of it.

More players needed
The handful of remaining big computer makers in the industry need to weigh in soon. Dell, Hewlett-Packard, IBM and Sun Microsystems survived round after round of consolidation in this industry that left HP as the world's largest technology companies by some measures.

Some of these companies may not survive the next big round of consolidation if the parallel breakthrough doesn't come and industry growth stalls. Despite that fact, I have not seen any of them show much leadership on this topic yet. I count ITRI, the research arm of Taiwan Inc., as one of the potentially players in this effort.

Many others have a lot at stake here including Google. The Wall Street darling might not be such an attractive whistle stop for the 2020 presidential campaigns if its data centers begin to bog down as they try to sort through growing piles of Web data and video.

Oracle needs to step into this spotlight as well. It has tried to consolidate the enterprise software world around its Redwood Shores headquarters. Now it needs to take a similarly sized stake in this industry problem.

It's not just a problem for mainstream computing. The parallel computing breakthrough will be needed to stoke the mobile Web that the cellular industry sees as its next big growth engine.

Plenty of people in the cellular industry have in-depth understanding about running software across multiple cores on handsets and big back-end base stations. They need to be part of the solution.

Someone also needs to take responsibility for mapping into mainstream computing all the good work on parallel programming going on in high-performance computing, especially the work in the HPCS program under DARPA.

There's plenty of room for plenty of stakeholders to weigh in on this problem with plenty of promising ideas. The only thing that does not exist in abundance is time.

- Rick Merritt
EE Times