Talk to a friend or colleague in IT, and pretty soon you’ll get on the topic of big data, the data explosions, and mega data centres.
Bar stool logic — and prevailing wisdom — suggests there is no end to our collective hunger for data, and so this must be fed by ever-larger, ever-denser data centres designed, powered and chilled in ever more innovative ways. Hyper-scaling is fun and interesting to discuss, and so we all do it.
Makers of servers, racking and networking equipment, power management and neat cooling doodads are keen to have us pursue this line of thought — to have us build out using the latest shiny, slim yet powerful gear that is filled with greater numbers of racks, packed with huge volumes of data and therefore demands ever more power. Well, maybe not.
European multinational Schneider Electric is bucking the trend of trying to squeeze in more power. Rather, this equipment and services supplier has been looking for ways that IT managers might find the sweet spot in terms of the power density of their equipment.
Schneider’s premise is simple enough — that rack power densities have a direct impact on the capital costs of any data centre.
So, rather than try to pack in more, perhaps a sweet spot of average rack density exists where maximum savings can be made before the additional infrastructure investments required exceed the efficiencies gained?
Wendy Torell, senior research analyst for Schneider Electric’s Data Center Science Center, explained: “We began research for a white paper because we were not convinced by all the forecasts claiming that densities would climb towards 20kW-30kW a rack and beyond. The only way to know for sure was to take a dive into why data centres in reality are still not at those densities, and what that might mean.”
Late last year Kevin Brown, vice president of the company’s global data centre strategy, gave a presentation entitled Don’t Predict the Future; Plan for it. He referenced predictions from industry observers including IDC and the Uptime Institute, which had published curves predicting average heat loads per rack reaching 25kW, 30kW even 40kW by 2009-2014.
He then countered this vision with a slew of media clippings highlighting the plight of data centres that had overbuilt as a direct result of flawed capacity planning, and were now left with oversized, underused facilities.
Brown claimed that the predictions were all wrong because loads shared across blades, servers, storage and network devices brought the overall average in any given facility down, while improvements in data technology and power management had given an inaccurate picture of the economics of density.
Also, most people overestimate the actual power consumption of their servers because they simply look at the boilerplate ratings rather than measuring them.