The Business Centre

Damoulakis says he believes that most people who ignore Moore's Law do so because its rewards -- at least in terms of storage -- seem invisible. While in theory total storage costs should come down as a benefit of the law in action, in fact, costs remain the same and go even higher in most environments, he says. The reason: Data-retention rates are outstripping Moore's Law. "Because you can keep more data, you are," Damoulakis points out.

The key issue, Damoulakis says, is maintaining a balance between data center components. And that's hardly the job of the enterprise IT purchasers. It belongs in the realm of the technology makers. When memory was expensive, systems were designed to make efficient use of it. When memory costs came down, the focus on efficiency dissipated. "You could almost argue that the drive toward server virtualization is because of excess computing power that exists and trying to find ways to leverage it more effectively," he says.

Through it all, Moore's Law has remained a steady guidance system foretelling the arrival of new technical capabilities -- whatever those might be. "There have been breakthroughs in so many points of time," Damoulakis says. "I think there's something kind of miraculous in all of it."

Building up, not in

Gregory Wong, an analyst at Forward Insights, which monitors semiconductor memory, agrees. "Every time we say there's a wall we're going to hit, it seems like [manufacturers] are able to rise up to the challenge and overcome those barriers," he says. As an example, Wong points to flash memory, which will one day use components as small as 32 nanometers. He wonders if that will be the wall. "Some people are saying it'll extend to 22 nanometers. Even if it goes to 22, then somebody will say, 'When we go to 15nm, we'll hit a wall.'"

The technical obstacles have to do with lithography -- the process of writing the circuits onto the wafer. "As these dimensions get smaller and smaller, it becomes harder and harder to make a transistor," Wong explains. The components become less reliable. New ways of building chips have begun coming to the forefront, including multicore chips that perform symmetrical processing and memory arrays that build up in layers instead of inward.

That may matter to chip vendors and systems integrators, Wong points out, but not to enterprise users. "If I'm an IT manager, I'm going to make plans for my IT department based on the strategy or needs of the company. 'I don't have enough space in the facility. Maybe I should move to blade servers, which is more space-efficient.'"

Doug Mechaber, a computer consultant, fits into that category. He says that rather than any direct application of Moore's Law at the companies where he has worked, budget and immediate growth plans dictate purchase plans for components such as servers. "Sometimes this results in underpowered servers, sometimes, overpowered," Mechaber says.

He prefers to plan for "do-overs," by choosing servers that allow for expansion of memory and CPUs. "I find the lifetime of most servers purchased this way is two 'Moore times' -- about three years," Mechaber says.