New Trajectories

Turning technology trends into breakaway products.

By Al Riske

1.May.06--To John Busch, the key to creating breakaway products is not intuition but analytical explorations.

A research director in Sun Microsystems Laboratories, Busch knows that's contrary to the way most of us tend to operate. "What most people do, they think they know the answer, then they build it, then they start to measure it and iterate from there," he says. "They start off and end up in a fairly narrow region of the design space and may be locally optimizing."

But what if you could accurately estimate the characteristics and key metrics of the whole system before you built it?

Soft-spoken but passionate, Busch has spent the past eight years in Sun Labs working with many of the best minds at Sun -- not to mention various universities -- to ensure that the company has the tools and techniques to do just that.

Of course the analysis capabilities and explorations have to be constantly rejuvenated with emerging architectures, but that's what keeps the job interesting.

One of Busch's favorite books is The Structure of Scientific Revolutions, by Thomas S. Kuhn, who coined the term "paradigm shift."

"Kuhn studied discontinuities in the history of science and found that new instruments and analysis capabilities drove many of them. Things like the telescope, the microscope, and mathematical models enabled new insights into what's really going on," he says.

"That's why we have focused so much on creating industry-leading analysis capabilities for computer system architectures -- and we have those now."

Those new capabilities, and the collaborative explorations that employed them, played an influential role in the development of the revolutionary UltraSPARC T1 microprocessor with CoolThreads Technology, our AMD Opteron system architectures, the ongoing High-Productivity Computing Systems (HPCS) project, and most recently new application server and clustered database architectures.

Beyond vision, Busch says an important part of his research role is acting as recruiter, collaborator, and enabler.

"I seek out the innovative thought leaders. When we do these explorations I recruit them in part time," he says. "I get the mix of people we need from universities and from across all of Sun's division. People like Marc Tremblay, Ivan Sutherland, Anders Landin, Kunle Olukotun, Rick Cattell, Brian O'Krafka, Dave Patterson, Robert Drost..."

"We pull together a broad system architecture team to explore the design space, and through detailed workload characterizations, proprietary modeling capabilities, and straw-man architectural analyses, we come up with a system architecture and high-level design that optimizes the key system metrics and puts us on new trajectories," Busch says. "The collaboration and analytical exploration are essential elements of our innovation and technology transfer model."

From these systems explorations come many surprises.

"Often what we expected to be secondary factors turn out to have first-order impact on the system metrics, while things we were focusing on initially turn out to be of secondary importance. During an exploration's iteration phase, key new ideas often come not from the people whose specialty it is, but from one of the other people just observing and thinking of some wild and crazy thing that stimulates the group. That's how innovation happens. It's messy, yeah? But when you have a microscope and a telescope, you can see a lot more of what is really going on. That's what our models let us see, all in the context of real workloads and alternative system architectures."

In the case of chip multiprocessing, modeling showed the new architecture would handle common Internet and database-related workloads several times faster than our best projected systems -- and be far easier to build because of the simplicity of the design.

Loath to take credit for any of this, Busch instead lauds both his Sun Labs team and the exploration teams. "Each member of my team is excellent, but together they are incredible in the capabilities and analyses they create," he says. "The system exploration teams do the architecture innovation."

Like many researchers, Busch cites the work of Harvard Business School professor Clayton Christensen, author of The Innovator's Dilemma and The Innovator's Solution.

"Christensen writes that established companies are rarely, if ever, successful with disruptive technologies, but what does drive their growth are radical sustaining innovations. Radical meaning new trajectories. Sustaining meaning along the current values and metrics," Busch notes.

In Sun's case, those values are price/performance, reliability, scalability, security, and so on.

"What I'm trying to do here is to create the engine to continually drive toward radical sustaining innovations at the total system level for each of our computer markets," Busch says.

"People don't understand well the difference between architecture, design, and implementation. They tend to focus on designs or implementations. Coming up with radical, innovative architectures along the primary values that our customers care about -- what I call breakaway trajectories -- that's the key to growth. Optimizing component designs and putting them together into systems has limited benefits."

What really gets him excited these days is a clustered database exploration his research team is working on with Sun's hardware and software product groups.

This exploration aims to exploit Sun's position as one of the few companies able to optimize the whole system -- hardware and software.

"Our approach helps us leverage that advantage in ways we haven't before," he says. "If you look where Microsoft is, they have the software, and Intel has the processor, and then Dell goes and puts things together using commodity parts. Each can only optimize their own little piece. But Sun actually controls all the technology levels, so we can do holistic system architectures.

"What we're looking at now," he adds, "is how to integrate the database and other software with advances in the switch fabric, the interconnect fabric, and the processors in very innovative ways. Sun uniquely has this opportunity. I expect this is going to be bigger than chip multiprocessing; this exploration has the same feel. "

In simple terms, Busch's team uses advanced statistical techniques to create very accurate component-level models, which they then combine into analytic system models for various architectural alternatives. They drive these models with representative workloads and estimate the key system and component-level metrics. In the process, the exploration team observes fundamental characteristics and trends. They also model the competitive context.

"So, early on, we can get very accurate, broad, and rapid explorations of the design space. We can see the impact of architectural variances and the contribution of each component on the system metrics, so we know immediately where the opportunities are," he says.

Busch notes that virtually every one of the system explorations Sun has done in this way has resulted in architectures and analysis capabilities that have had a big impact on the company, its products, and follow-on design processes.