Networks are sucking power: should you worry?


Power efficiency hax become a major factor for data centre equipment. While vendors compete on greenwash, users need to test their actual power usage

Most of the publicity concerning power and cooling issues in data centres has tended to focus on how efficient computing equipment is. In contrast networking gear has tended to take a back seat.

But as networks based on 10 Gigabit Ethernet become prevalent, as they inexorably are, power consumption will rise, and power usage, cooling and power efficiency will become entries on the data centre manager’s worry list, according to test equipment vendor Ixia.

There are standards initiatives, with the IEEE producing “Green Ethernet “. Several vendors have started marketing around their lower power switches – eWEEK has spoken to 3ComBroadcom, and Extreme.

But the issue is actually in the end user’s hands, says Ixia. What’s needed is to reach a common measurement standard – and to test against it.

Electricity prices up by 83 percent

Routers and switches are getting denser, and power consumption is increasing faster than either cooling or developments in new silicon can reduce it. So companies’ operational expenditure “is non-trivial”, says Michael Haugh, senior product manager of Ixia’s infrastructure division, not least because the cost of electricity has increased by 83 percent since 2003.

“The network market was not as power aware as the servers and computing market, but it is now. It used to be just build out and deliver the bandwidth,” says Haugh. “The energy effect was an afterthought in the network infrastructure business. We had to do planning and cooling when designing data centres when I worked for IBM, but never did the criteria involve energy efficiency.”

So we need a new way of measuring the performance of network equipment in the data centre. The problem is that “there’s no standard in the data centre for energy efficiency. To do this means understanding how you rate the device and, as each device has different capabilities, you need a way of dividing equipment into classes to make them comparable,” says Haugh.

PUE doesn’t work for networks

The metric the industry seems to be settling on is power usage effectiveness (PUE) but, while PUE is a good metric for computing devices such as servers, it’s not appropriate for network equipment, says Haugh. “Telecoms people aren’t asking for it.”

“For sophisticated devices you need unique methodologies for measuring and providing a rating at different points. For example, you need to measure power use at idle, 50 percent and 100 percent throughput and look at the performance spectrum across those usages.”