You can hardly pick up a business or IT publication these days without finding someone exhorting Hewlett-Packard to "reinvent" itself.
Regardless of how, or if, new CEO Mark Hurd does that, IT seems likely to go on quietly reinventing itself inside HP Laboratories. The labs may get only 5 percent of HP's total research and development budget, but they're working on a broad array of technologies, from data centre management tools that are expected to find commercial applications next year, to new computer architectures that won't hit the marketplace for at least seven years, if ever.
"We try to be out in front of the company," says Robert Waites, director of strategic planning at HP Labs in California. "We try to skate to where the hockey puck will be, not where it is today."
Many of HP Labs' 700 employees are now skating toward a "reinvention of the economics of IT", one of six broad research areas that includes projects in grid and utility computing, self-managing systems, virtualization and smart data centers.
"The most fruitful places to innovate are now above the commodity operating system and CPU chips," Waites says. "We have very little work going on in CPU architectures, but 20 years ago, that was a dominant research program."
What's in store near-term
Beth Keer, manager of storage systems research, says most IT shops spend 80 percent of their budgets on hardware and software maintenance. The goal of a suite of projects at HP Labs is to knock that down by almost half. The key is to automate IT tasks such as provisioning disk arrays and configuring networks, she says.
"There are many steps, and if you screw it up, you are in big trouble. And because these tasks are repetitive and complex, they are not a good fit for human cognitive skills," Keer says.
Projects that attack this problem lie in two broad areas: virtualization, and automated management and control. They include the following:
- SoftUDC. The software-based Utility Data Center is a prototype tool for virtualizing server, network and storage resources. It creates a logical layer across disparate hardware and a single, centrally managed pool of resources.
- FAB. The Federated Array of Bricks consists of low-cost, industry-standard hardware and proprietary software that allows easy provisioning of storage systems. A "brick" holds a number of disks and a CPU controller. Additional bricks can be snapped in for "capacity on demand", with the Linux-based software automatically striping data across the bricks and providing for redundancy in case of failure.
- SLIC. Statistical Learning, Inference and Control tools use pattern recognition and probabilistic models to identify aberrant system behavior. Research is now focusing on forecasting problems.
- Smart Data Center. This project involves figuring out how to better cool ultra-dense components such as blade servers while saving on energy costs. "Dynamic smart cooling" uses thermal modelling, networked sensors and even robots to lower cooling costs by 70 percent, HP claims. Keer seems undeterred by the technical challenges in her work, but she acknowledges some doubts on the user front. "There are some human factors about people's reluctance to adopt new technologies," Keer says. "If they can't see what's going on, do they trust the automation?"
While Keer works on things that have one foot in the marketplace, HP Labs' Duncan Stewart is focused on something unlikely to have any payoff for seven to 10 years. The research physicist and his colleagues are hoping to shrink computers to almost unimaginably tiny dimensions.
For more than six years, HP Labs has been inventing a radical new approach to computing based on crossbar technology. HP's crossbars are molecular-scale circuits consisting of grids of wires whose intersections can be populated, by programming, with various devices such as resistors, diodes and switches. Several years ago, HP showed that these crossbar arrays could be used to make memory and very simple logic circuits far smaller than equivalent circuits made from silicon transistors.
But HP found two show-stoppers on the way to making a practical computer: there seemed to be no way to restore degraded signals as they travelled from one logic gate to another and no way to do signal inversion, which is necessary to perform the Boolean NOT operation. Both functions are a cinch with silicon transistors.
Then, in February, HP Labs announced a breakthrough -- a way to perform both signal restoration and inversion using a pair of very simple molecular-scale switches combined into a crossbar latch.
"Latches are the glue that holds together all of the different pieces of memory and logic inside a processor," Stewart says. "That was the missing piece that will enable all kinds of computing to be done at the molecular scale. We are going to build the smallest computer in the world."
Meanwhile, conventional chips will become extremely difficult and expensive to make as they get smaller. A published roadmap for the semiconductor industry has the smallest distances between wires on a memory chip shrinking from 90 nanometers today to 65nm in 2007, to 45nm in 2010, to 32nm in 2013 and on down from there.
"What they are going to do 12 years from now is mapped out, but they don't have a clue how to do that," says Stewart. "In fact, they think they may not be able to do it."
The 32nm milestone is "a reasonable place for us to inject some of these ideas," he says. The idea isn't to replace silicon transistors but to build certain devices, such as ultra-dense memories, on top of CMOS chips. Stewart says HP hopes to eventually build crossbar devices smaller than 3nm. Meyya Meyyappan, director of the Centre for Nanotechnology at NASA's Ames Research Centre, says it's too early to say whether HP will succeed. "Until today, everyone was doing straightforward silicon CMOS-like technology," he says. "As such, there was nothing novel. But the crossbar architecture is a novel concept with the potential to lead towards future-generation electronics."
One application of these Lilliputian computers might be to give tiny sensors, or "motes", enough processing power to perform very compute-intensive functions. For example, Stewart says, "If I can deliver you a very small computer -- a few microns square -- that can run on power it soaks up from the environment, then things like RFID tags can have cryptography."
Could there be more show-stoppers? "The biggest one I've seen in research labs is economics," Stewart says, after some thought. "When your technology is actually ready to go, the market may not be ready for it."