Cloud Computing: Power Hog or Panacea?

 [This article was originally published in Enterprise Conversation, a UBM/DeusM publication. ]

As 2012 passes into the bit bucket of history, and the world once again fails to end on cue, a new litany of unfulfilled dire consequences is emerging.  Back in 2006, the trend toward data center consolidation was first taking off and what we now call “cloud computing” was first catching on.  At that time, the U.S. Environmental Protection Agency estimated that within five years, the rapid growth in power demand by data centers would lead to a doubling of the number of kilowatt/hours (kWh) that power stations produced.

The End

A few years ago, I compared the EPA’s numbers to the Dept. of Energy’s electricity delivery estimates, based on the rate of population growth in the U.S. Census.  I concluded, and later independent estimates confirmed, that if the EPA’s estimates were accurate, the amount of supplemental power draw required by new data centers would be greater than the power draw required by new people — new members of the U.S. population — in 2011 and beyond.

The cost of this supplemental power draw was left, as with so many other topics that are too unpleasant to be thought about, as an exercise for the reader.  Since I was up for the exercise, I calculated the budget cost per year for U.S. government contributions to the construction of new power stations just to satisfy this new demand.  I came up with $21 billion per year, which was 56 percent greater than the entire per-annum discretionary resources budget for the Dept. of Transportation.

A New Beginning

In the intervening years, independent analyses of the efficiency benefits gained from cloud computing told absolutely the opposite story: not catastrophe, but rather paradise.  Citing the very same Environmental Protection Agency, consulting firms produced evidence of tremendous carbon emission reduction on the part of businesses that had shifted their application workloads to Microsoft cloud-hosted SaaS.  A 2010 Accenture study (PDF available here) revealed that carbon dioxide emissions among users of Microsoft Exchange were reduced by 52 percent per user among large enterprises, and over 90 percent among SMBs, simply when they switched to Hosted Exchange from on-premise.

The conclusion Accenture drew:  Lower carbon emissions comes from more efficient use of power, as a result of using less of it.  “Generally speaking, the comparatively smaller carbon footprint of cloud computing is a consequence of both improved infrastructure efficiency,” the report read, “and a reduced need for IT infrastructure to support a given user base,” before going on to directly link data center efficiency to power usage effectiveness.

The following year, in its contribution to the Carbon Disclosure Project, Verdantix used the EPA’s carbon reduction numbers to project energy savings among some 2,653 U.S. firms whose annual revenues were $1 billion or above.  The firm projected the combined net total energy cost savings for these firms in 2013 to be around $2.5 million per annum, rising to about $12.3 million per annum in 2020 (probably also accounting for rising energy prices), simply for having moved to cloud services delivery models.

So in seven years’ time, conceivably, about half the annual cost of producing new power to meet growing demand for large power centers would be offset, if these predictions hold true, by reduced power demand among general businesses.  The suggestion is that, within our lifetimes, it could all come out in the wash.

A New End

Not exactly.  As anyone who builds his own PCs knows, the real power drain comes not from the processor, but from the cooling system.  In fact, thanks so very much to the laws of physics at the subatomic level, the smaller a processor is scaled and the less power it draws, the more power may be required to keep it cool, unless you innovate the way it’s cooled (thus, the tablet PC).

A 2010 General Services Administration report (PDF available here) made the clear case that the cost of cooling systems was rising faster than not only the cost of using them, but even of acquiring them in the first place.  The year in which cooling the data center would become more expensive than owning a data center: 2012.

But it didn’t stop there.  Because of the way the nation’s energy infrastructure is delivered, it went on, for every 33 units of energy that a data center actually uses, another 66 units was wasted actually getting it there from the generator.  That calculation suggests that the offset in energy savings for corporations from processing efficiencies is more than offset (perhaps doubly) by the catapulting costs of our energy infrastructure.

So did this catastrophic confluence of predicted events actually happen?  What is perhaps most astonishing of all is that we don’t know; it would appear that no one — not the GSA, EPA, DOE, DOT, PBS — has bothered to check.  Not being able to grasp the scope of the problem in front of us may be becoming the defining characteristic of America in the 21st century.