Breaking Analysis: The NY Times article published over the weekend on datacenter power is a rehash of old data and old concepts that no longer apply to today’s datacenter market. Another one is up today called Data Barns in Farm Town, Gobbling Power and Flexing Muscle.
The NYT article has pulled together a lot of unrelated industry numbers in conjunction with vague, random numbers from McKinsey to shine a light on datacenter inefficiencies.
We have an angle on this data center story because SiliconANGLE.com and Wikibon.org have in-depth coverage this area (and enterprise), so here is our opinion and data on this new Cloud Series by the NY Times. (Video analysis below.)
This is a publishing and reputation fail by the New York Times.
Two main reasons:
1) it took one year to write (that makes them look stupid) and,
2) the entire cloud series is more doom and gloom than anything else.
Bottom line: the entire New York Times’ article and current cloud series is suspect. The New York Times focus on the datacenter is not only irrelevant to the direction of modern society, but their entire article is filled with inaccuracies and old data.
Unfortunately, what the story doesn’t show is how dramatically the curve on real, substantial efficiencies has risen over the last 5 years.
Today’s datacenter revolution is still about availability
Companies who have datacenters care about one thing: Availability. This hasn’t changed in the entire history of Information Technology (IT). Over the past decade our world has seen, and is experiencing today, a revolution in modern services never seen before. Hello iPod and iPhone. Hello Android devices, the Xbox, App Store and YouTube. Hello cloud computing, mobile, and so much more. There’s so many technology advances and solutions coming down the pike.
Datacenter and IT personnel have been getting smarter because they are solving these new large-scale, modern problems and building new solutions that one has ever faced before. Education for these new, modern areas do not exist and the challenges of running a modern day infrastructure and datacenters requires new knowledge and capability of running large scale software, hardware systems. It’s about availability.
Gone are the days of how much square feet do we need – that was so year 2000. Today in the datacenter only one conversation matters world – availability and how many kilowatts do we need and how much. Everything else flows from there.
With respect to the power consumption problem/issue, as it was pointed out in NYT, there is nothing different in datacenters that differ from any other “thing”. The physics of electricity is the same for datacenters as it is for all other “things” like manufacturing, service businesses, physical factories, homes, and businesses.
Datacenter innovation advancement, not pollution.
Here is my response to the NY Times article. The article failed miserably at covering just how much advancement have been made in driving datacenter efficiencies over the last several years and highlighting new technologies that have made a real difference for real datacenter operators.
More than two-thirds of all applications are virtualized today, meaning that they likely run on a single server with 8-10 other applications. Five years ago, each application would have had its own server. So new applications deployed today essentially require 1/10th the resource of previous generations of technology. Server virtualization alone took the market from growing at double digit power consumption in 1997 to low single digit growth today.
Most major corporations have undertaken substantial datacenter site consolidation projects. The most visible right now is the US government has goal to reduce the number of their datacenters. This process of replacing old buildings and inefficient M&E equipment with new, state of the art facilities where power and cooling is the building block of design. Many new datacenters are modular in design, meaning that they are built as capacity requires, not overbuilt from day one, as older generations had been.
The introduction of datacenter automation software will continue to drive substantial improvements in utilization levels of power related infrastructure. We regularly see power consumption improvements of more than 25% among companies that deploy DCIM software where server, storage and M&E equipment are linked via software specifically to optimize the performance of all systems together. This market is still emerging today, but is maturing quickly and will be a cornerstone for future datacenter design.
Availability, not power consumption
There is only one thing important in a datacenter – power availability — not power consumption. Kilowatts as a metric is the blueprint for every datacenter built as operators want to ensure that given limited power available, they can support the maximum number of systems. Without power, a datacenter is out of business. Datacenter operators and designers will conserve energy only if they think they can extend the life of a datacenter, not for any sentiment around being green or environmentally friendly. Sorry, those are the hard facts.
Regulation is an option, but these regulations are usually pulled together by a strange assortment of governmental agencies, engineers and industry constituents with different agendas. This usually leads to a set of sometimes-related rules voted on by compromise that may or may not lead to optimal datacenter design. And really, given that the US government has the worst, oldest, most poorly designed datacenters, do we really want them dictating the future datacenter? Please, no way.
The New York Times needs to get it right
What works are incentives, money talks bullshit walks. Given the reach (circulation) of the NY Times in rallying attention, I really wish this article had spent time highlighting success stories and new solutions that work. Lets have our government spend time and attention in supporting R&D initiatives and rebates, not fining the local Amazon datacenter for being out of compliance on permitting of their diesel generators
Finally, there is one number in the article we should focus on. Koomey estimates that less than 2% of all electrical power in the US goes to datacenters. The other 98%+ is for heavy industry (such as manufacturing), hospitals and homes. As a voter, I’d much rather spend my tax dollars fixing the real problem, not trying to slow down one of the few bright spots in the economy today – the growth of the digital industry.