Virtualisation cuts carbon and servers

Now virtualisation is mainstream, fears about putting too many apps on the same server are evaporating. That just leaves the problem of calculating power consumption, as Stewart Baines reports.

Computing may drive efficiency but the data centres that run business are often far from efficient. Servers haven't traditionally used all of their spare CPU power, because reliability concerns stop administrators running lots of applications on one box.

The extra energy used for those wasted CPU cycles means that an enormous amount of carbon is burned into the atmosphere unnecessarily. But how much? And more importantly, how much energy - and therefore carbon emissions - can be saved by consolidating servers?

Fujitsu Siemens Computers recently released a report suggesting that the UK's data centres are wasting enough energy to burn through 1.2 million tonnes more carbon than is needed.

A report by the US Environmental Protection Agency in August suggested that simply by continuing current rates of server consolidation in the data centre, the US could save 15 million metric tonnes of carbon by 2011.

Working out a company's potential carbon savings from virtualising servers requires a before-and-after calculation of energy usage. That also happens to be a useful metric for those trying to quantify energy savings. So how are those calculations made?

"It depends on what hardware you're consolidating," says Martin Niemer, senior product marketing manager for EMEA at virtualisation pioneer VMware, who argues that IT departments will be able to squeeze more productive CPU cycles from one box than another, depending on what is running, how many cores are in the process and how much RAM it has.

The firm has capacity planning tools that partners can use to estimate the energy savings. "But if you're a smaller customer and just want a general understanding, then the rule of thumb should be a consolidation ratio of one to 10."

Rules of thumb can often be over-simplistic. The new, consolidated server will consume more power than one of the 10 that it replaced, for example. Other parameters include heat production and, consequently, the energy invested in data centre cooling. No wonder that shoring up the case for consolidation with numbers to suit your exact application profile and hardware is a challenge.

Nevertheless, some have taken a stab at it. For example, outsourcing consultancy Atos Origin built a power consumption and carbon production model for a representative data centre, explains chief technology officer for managed operations Guy Lidbetter. The model, although informal and unaudited, has served the firm well.

"Our representative data centre provides 30,000sq ft of floor space and hosts 3,000 servers of varying type, form factor and operating system. This has been assessed, at current typical utilisation and efficiency levels, to produce 5,540 tonnes of CO2 per annum at a cost of £1.159m for the power consumed," Lidbetter says.

Adjusting the baseline to reflect projected energy price increases in the next three years increased the annual cost of energy by two-thirds, but the numbers suggest that virtualisation would claw back that cost and more.

A what-if scenario that increased server CPU use from 15 per cent to 60 per cent slashed the amount of energy used and therefore reduced overall power costs by 87 per cent, Atos found, despite the increased cost per kilowatt.

Those without the resources to roll their own cost models might want to use someone else's. Nick Carr, marketing manager at open source software firm Red Hat, points to the True TCO Calculator spreadsheet at the Uptime Institute, which includes multiple criteria affecting the ultimate kilowatt hour (kWh) count that will dictate energy costs. Other resources include the cost calculator at server consolidation specialist PlateSpin.

Or you could virtualise first and answer questions later. That's what Chris Rogers, CEO and president of ISP and hosting firm ServerCave, did. The company ran 100 servers across three racks, consuming a total 12kW, he says.

He used Xen virtualisation software to reduce the hundred Linux-based servers, which used direct-attached storage, down to 10 attached to a SAN. Consolidation reduced total server power consumption to 3kW, and another 1kW for the SAN, cutting 75 per cent of ServerCave's energy use. At an electricity cost of 19 cents per kilowatt hour for 24/7 operation, that saves $13,000 per year.

At the end of the day, however, energy cost saving calculations may not be that important to computing departments, most of which don't pay their own electricity bills in the first place. Such burdens are left to the facilities management team, who may be in a different part of the country, managing energy costs on a much broader scale.

Nevertheless, the saving in kilowatt hours gives companies a useful starting point for a conversation with its energy provider over carbon emissions - and that's something that will be of interest to anyone concerned with corporate social responsibility.