Foster D. Hinshaw knew there had to be
a better way. His clients with large database systems were in
trouble, and the situation was getting worse. Some of the biggest
organizations, including the Commonwealth of Massachusetts, were
finding that BI (business intelligence)
wasn't so useful if it couldn't handle all of their data.
Hinshaw's better way eventually led to
a whole new way of handling data that uses the specialized database
appliances built by Netezza, the company he helped create in
2000.
"I realized it was a data-flow
problem. You had to change the way the data flowed," he says. So
Hinshaw gave it some thought. "One of the things I do very well is
visualize data flow," he says.
The more Hinshaw thought about the
data flow in a typical BI system, the more he compared it to the way
air flows. You can have smooth airflow, called laminar flow, and you
can have turbulence. "There are a lot of places where software
becomes turbulent," Hinshaw explains. "But to do BI, you need smooth
data flow."
So how is it that Hinshaw is using
aerodynamics to design databases? His varied background ranges from
designing electric cars to designing real-time operating systems and
computing hardware. That experience has led him to realize that the
only way to faster and more effective database operation is through
design changes.
To get the smooth laminar data flow he
wanted, Hinshaw understood that the data entering the flow had to be
restricted to just the data he needed, not all of the information in
the database. This meant that most of the initial database work had
to be done at the source, so he started by designing intelligent
storage devices that would extract only the data needed and send it
to the computer's CPU in a steady stream of information, in what
Hinshaw calls predigested results.
But that's only part of the work. Once
it arrives at the CPU, it's broken down into pieces, called
snippets, and is processed in what Hinshaw calls an asymmetric
massively parallel processing unit. This CPU is designed so that
information can move quickly.
"It doesn't matter how many CPUs you
have, but rather the performance of the fabric and the backplane,"
Hinshaw says.
Put it all together, and the results
are impressive. The company's performance tests show that the
Netezza Performance Server 8000 is producing results in seconds that
comparable SQL platforms take days to produce. Apparently, there was
a better way.
"What’s unique about this is that
there have been prior efforts to do some of this, but we’ve done the
best of both worlds," Hinshaw says.
(For profiles of the other nine 2003
InfoWorld Innovators, see Honoring the
Innovators.)