Moore's Law can't stand the heat

In this special report, we look at why many datacentres today are facing a power and cooling crisis.

Crisis in the Datacentre

Over the past few years, the amount of electricity required to power a server in a datacentre has more than doubled. So although buying the server costs almost 20 percent less than it did two years ago, that server costs significantly more to run.

This increase in operating costs is because the latest servers are stuffed with more computing horsepower than ever before. Even though the chip manufacturers are creating more efficient processors, technologies such as virtualisation are causing these chips to work harder and for longer periods of time.

The hidden cost here is that, with more transistors in each processor and more processors (and processing cores) in each server, the datacentre rack generates far more heat than it used to, which means traditional datacentre cooling systems can no longer keep up.

The problem is so bad that analyst firm Gartner expects to see more money being spent this year on power and cooling technologies than is spent on the actual servers themselves. In two years time, Gartner expects around half of datacentres to have insufficient power or cooling capacity to meet demand.

The cost of powering and cooling a server is currently 1.5 times the cost of purchasing the actual server itself, according to The Uptime Institute (TUI), which is a think tank dedicated to improving datacentre reliability.

In a white paper titled Data Centre Energy Efficiency and Productivity, Kenneth Brill, founder and executive director of TUI, said that if current trends continue till 2012, best-case estimates show that powering and cooling a server will cost three times as much as purchasing the hardware. The worse case scenario is that power and cooling will cost 22 times more than the hardware.

This prediction has led the TUI to go as far as saying the benefits gained from Moore's Law -- which states that the number of transistors crammed into silicon will double every 24 months -- will be nullified by the increasing cost of powering and cooling those processors.

In a recent article for, Brill argued that Moore's Law can no longer be seen as a "good predictor of IT productivity because rising facility costs have fundamentally changed the economics of running a data centre".

Brill gives an example where one of TUI's members spent US$22 million buying new blade servers but then had to fork out US$54 million to boost power and cooling capacity. This increased the return on investment figure from US$22 million to US$76 million.

He warns CIOs that their "facilities and infrastructure" costs, which currently account for between one and three percent of the IT budget, are likely to shoot up to between five and 15 percent within a few years.

"That is enough for the CEO and CFO to begin scrutinising how the IT budget is being spent," he said.

Bill Clifford, chief executive of datacentre management software firm Aperture, believes there is an impending crisis: "Many users are simply unaware of the dangers they are introducing to their datacentres".

Clifford said that in the past, cooling issues could be resolved in a matter of weeks as suppliers of cooling equipment could cope with demand. But with the increased demand, the lead time for new equipment can be up to 18 months, putting firms at risk of experiencing datacentre downtime.

Virtualisation is the silent enemy
Virtualisation has been touted as a technology that can be used to reduce power consumption because it allows computing tasks to be consolidated to fewer servers. Unfortunately, reducing the number of physical servers by increasing the workload on the remaining servers can result in increased power consumption.

According to Kris Kumar, MD of Sydney-based datacentre design specialists 3iGroup, virtualisation is the "silent enemy" because once you virtualise a server, it is likely to run at about 80 percent of its capacity. Before virtualisation, that server was running at about 15-20 percent of capacity -- and therefore using less power.

"In real power terms, a 300 watt server which was running at 20 watts is actually now running at 280 watts. You are reducing the footprint but putting in more processing power, so the power per footprint has gone up," Kumar told ZDNet Australia.

Tim Ferguson from contributed to this report