Last week’s Verdantix and AT&T report on the energy and carbon emission savings to come from cloud computing is the latest in a long line of studies stating the obvious: Shifting computing from inefficient, dispersed data centers to highly efficient, centralized cloud data centers should save everyone energy. Indeed, the report’s claim of $12.3 billion in energy savings over the next decade fits in with other figures, such as Pike Research’s prediction of a 38 percent energy reduction from a wholesale move to the cloud or Microsoft’s claim of 90 to 32 percent energy reductions from its cloud-computing offerings, depending on how energy-efficient the customer’s deferred data-center investment would have been.
But when it comes to energy, what cloud providers and customers really care about is controlling data center costs and avoiding unneeded investment. To try to get at the cloud’s energy benefits, the industry will need transparency that can allow market forces to determine just how valuable those energy savings truly are — and that means data-center-efficiency metrics. There are several measurements you will need to know to figure out how to differentiate green-cloud-marketing hype from reality.
The data center writ large. So far, most green data center PR has been focused around power usage effectiveness (PUE) and data center infrastructure efficiency (DCIE). These measure a data center’s overall efficiency at using power to support its core IT assets, though from different angles. The Uptime Institute’s corporate average data center efficiency (CADE) measures utilization of both IT and data center facility assets across the enterprise. All of these stats get at how efficient a data center operator is at managing the non-IT energy costs — cooling, lighting, power delivery and backup — compared to the “working” IT power load.
No doubt, cloud-computing providers can offer far better PUE, DCIE and CADE measures than the data centers they’ll replace. Just compare the 2.0 PUE industry average to hyper-efficient cloud data centers from Facebook, Google, Cisco, Amazon and Microsoft with PUEs that approach the perfect figure of 1.0.
Power at the server level. But most cloud-services clients’ needs will be counted on a by-the-server basis. How to compare one server with another in terms of the efficiency of its operations? Getting at that figure is harder than just looking at a server’s specifications, since it also involves variables like how much the server is being used.
Still, a number of statistics are seeking to deliver some kind of computing-power-per-watt measurement. The nonprofit Standard Performance Evaluation Corp. has its SPECpower metric, which measures server performance in operations per watt. Data-center-efficiency-technology startup Power Assure has released its own version, called PAR4, which measures server power use at usages from idle to peak power, as well as on a transactions-per-watt basis. Deutsche Bank has developed a set of metrics that involves comparing a company’s existing servers to the best available to come up with a figure called hardware utilization efficiency (HUE).
With all of these measurements, some standardization is in order. The Green Grid — the industry group that came up with PUE, as well as measures on carbon emission and water-use efficiency of data centers — is working on a metric called data center compute efficiency (DCcE). It’s also working on a server compute efficiency (ScE) measure, which would go into calculating an entire data center’s DCcE.
That could give the industry some much-needed transparency into how to compare servers to one another in both dedicated and virtualized environments. Lost in much of the debate over the cloud’s energy savings is how the cloud will be priced into the equation. After all, every dollar of energy savings to come from switching to the cloud will be split between the cloud provider and the cloud customer. If the cloud host is saving a ton on energy, though, it’s not likely to let that margin go, unless it’s forced to in a pricing war. But we might have to wait for cloud services to get more popular before we see market data to prove that these metrics matter.