We certainly have plenty of problems to solve in the datacenters of 2012–but nothing like what James Glanz of the New York Times portrayed last week in “Power, Pollution and the Internet” (PPI).
The most sense I’ve made of this article is to read it as an investigative assignment that turned into a lifestyle piece. The Times apparently invested a year into thinking about environmental degradation associated with computing, didn’t know what to do with its findings, and reshaped them to fit a more familiar narrative model. People think computing is hip and cool, Glanz tells us, but–lo, the dramatic tension–in fact it requires much electricity generated in nasty old power plants.
True enough. When Glanz attempts to explain the detailed engineering behind data centers, though, it all turns wrong, or, more precisely, it turns into the kinds of explanation most often found in public-school textbooks: the words are all ones the technically-adept use, but they’re out of order.
Plenty of knowledgeable commentators have already called out the errors and confusions in PPI. Charles Babcock, who has been making difficult IT subjects clear for decades, shows how “N.Y. Times Data Center Indictment Misses Big Picture” in Information Week. Bob McMillan, another deeply-experienced reporter, tells “The New York Times Says Data Centers Suck–But Here’s What They Missed” for Wired. Forbes is particularly detailed in its “Response to the NY Times: Datacenter Is About Innovation Advancement Not About Power and Pollution“. While I think author John Furrier overemphasizes “availability” as a motive, he’s on-target in summarizing that “[t]he article [PPI] failed miserably at covering just how much advancement have been made in driving data center efficiencies over the last several years …”
Kevin Roose puts PPI in appropriate journalistic context for New York Magazine in Geeks Cry Foul at the New York Times‘ ‘Big Data’ Series“: it smells like “too many editors wrangling too many ancillary sub-narratives into a finished story.” Hollywood already had a slogan for the risk in 1981’s “Absence of Malice“: “accurate, not true”.
From a systems-and-software perspective, Sharon Fisher specifies “3 Things the New York Times Data Center Story Left Out for IT Knowledge Exchange”.
I usually avoid crowd scenes (be aware that there are dozens of other more-or-less incisive responses from Gigaom, CNet, Slashdot, and so on) like this; I’m more valuable in quieter locales. There are so many confusions in PPI, though, that a few technical points that I think remain relatively neglected in all this deserve mention:
A fundamental principle of both economics and engineering is to evaluate decisions with respect to alternatives. This is the most discordant omission in PPI: while the article is full of characterizations of “waste” and “inefficient use”, it essentially never structures its inquiry in terms of realistic choices. Is there something about data center management that makes “waste” profitable? What is the efficiency of computing done outside the data center? Does the author understand that many situations call for a switch to diesel to reduce exhausts? How many architects are still using old RAID models inappropriate to highly-virtualized workloads?
The DevOps bottom line: if your civilian friends try to talk to you about the dirty business you’re in, assure them it was all a big misunderstanding. Keep working on the real problems of security, availability, recovery, and operational efficiency already before you. The solutions–solutions in terms of engineering systems which fulfill business requirements–you create within a traditional datacenter framework will also go a long way toward making the most of the electricity, pollution, and other costs contingent on computing. We’re getting better at this every year, and none of the evidence the Times presents deserves more attention.
Come back on Thursday of this week, when the “Real User Monitoring blog” looks at a security problem that’s real, serious, and difficult.