Sunday, May 31, 2009

Old habits die hard: the bad news on data center energy efficiency

Despite the batch of pretty good news I reported in my previous post about the trends we see in how data center managers are approaching energy efficiency from 2008 to 2009, there is some bad news. Isn't there always?

But before we get too gloomy, it should be noted that much of the media, vendor, and customer discussion about energy efficiency over the past few years seems to have paid off in getting the word out. As I discussed last time, more IT folks have "green" initiatives to leverage, and more are measuring their power consumption. Sure, that sometimes means that what they measure is pretty inefficient or problematic ("Um, you know we're out of power in our New York data center?"), but at least they're measuring.

The conversation has led many (including my own esteemed colleague and Cassatt chief scientist Steve Oberlin) to point out that the current cloud computing discussions can absolutely be seen as one way that data centers -- or at least the organizations that own them -- can become more green.

The "green IT" messages just might be getting through, but...

All this discussion seems to mean that organizations like the EPA's Energy Star program, the Uptime Institute, the Green Grid, and others are getting through to people who run data centers. For example, in this year's Cassatt Data Center Survey, slightly more (63.2% v. 61.4% previously) know that the EPA recommends turning off servers when they aren't in use. Even though it's just a slight gain, I'll take it.

However, our survey also uncovered enough backsliding from 2008 to 2009 that I'm still forced to point out that some old habits die hard. Even if some of those habits are hurting the operation of your data center. In fact, some of the concepts that Cassatt has been strongly advocating, are meeting some very stiff opposition. Chalk up a few points for operational inertia.

Example: many would consider shutting off idle servers, but the percentage has dropped

Here's an example of one case where the "business-as-usual" approach is still holding on pretty firmly: one of the simplest use cases for Cassatt Active Response has been to use our software to shut down servers when they become idle and then use our policy-based automation to turn them back on when needed. However, powering up and down servers has long been seen as a no-no in IT operations, despite a wave of sources (including the EPA and the Green Grid) recently advocating this approach.

Given the long-standing skepticism about fiddling with server power, our 2009 survey's result actually seems impressive: 55% of the folks who responded to our survey (IT operations, data center managers, and the like) could justify turning off servers. Seems significant, right? Well, it is, actually, if you have a good feel for the underlying conservatism of those charged with keeping the data center running. The sad news from our perspective, however, is that this number is down slightly from last year's figure (59%), showing that the entrenched management ideas are still very strong. That’s despite some pretty good savings estimates. (Our savings calculators conservatively show Active Power Management-driven cost reductions starting at around 23%, with potential for closer to 50% in many cases.)

Similarly, the "deal breakers" for server power management remain similar to what they were last year: application availability is the most important. Impact on physical reliability and application stability were tied for the 2nd most important. This year's numbers do show a surge in worries about the physical reliability of machines (36.3% to 42.5%) and in the potential application downtime that people perceive server power management might cause (45.3% to 51.4%). But if you end up with a heat-induced outage like had today, suddenly some proactive server shut-downs to avert a literal data center meltdown may not seem so scary.

So, any signs that the conservatism in IT operations groups can change?

Actually, yes. Respondents seemed to have grown more comfortable with determining ROI around the topic of server power management. (And, no, I don't think our aforementioned savings calculators can be credited with that.) Also, despite an increase in skepticism regarding using automation for server power management, 36.7% still said they would be OK using automation to power manage a majority of servers in their dev/test environments -- exactly the kind of advice we have been giving prospective customers. Interestingly, 27.2% even said they'd do this for low priority production servers.

Though the IT/facilities gap remains, it is shrinking

By the way, one of the issues alluded to in nearly all writings on the topic of data center energy efficiency -- the alignment gap between IT and facilities -- is still there. But the gap is closing. When asked how integrated facilities and IT planning are, 29.7% said there was "no gap" and they were "tightly aligned," 37.0% said there was a "small gap" and that they speak with their counterparts in the other organization "somewhat."

Where was the improvement? Last year, 32.2% said they had either a "significant gap" in which IT and facilities touch base "infrequently" or a "large gap," meaning they "don't interact at all." This year those numbers dropped to 20.3%. Maybe these organizations are being brought together by smart companies looking for answers to their data center energy problems. Or maybe these guys are just taking the advice of Ken Brill of the Uptime Institute or various analysts and doing something as simple as taking their IT and facilities counterparts to lunch. I'm happy either way. It's amazing what a little communication can do.

So, are the 'experts' succeeding in being heard about data center energy efficiency?

One of the odd things we noticed in last year's survey was that, despite there being a great deal of independent, unbiased expert advice out there regarding data center energy efficiency, respondents got most of their information on the topic from entrenched system and power/cooling vendors. You know, the ones with a big stake in keeping the status quo. (Of course, it could be argued that these vendors have a good perspective on what's needed. However, these vendors just don't have the economic incentives to push radical change.)

And 2009? Same thing. Expert bloggers and media websites on the topic (like TechTarget's SearchDataCenter and others) did get significant mention. Peers and industry analysts did well, too. But the big guys are still the ones that folks are going to for guidance.

There was a curveball this year, too, however. The Uptime Institute, the Green Grid, the EPA, and even the Silicon Valley Leadership Group (which put on a great event about "green IT" last June) all did unexpectedly worse than last year when asked where folks go for data center energy efficiency guidance. Hmmmm.

Big problems take a long time to fix, so don’t expect instant improvement

The take-away from all this? I look at it this way: the data center power problem is a big, long-term one. The processes and approaches that have gotten us all into this situation are big, long-term ones. And, therefore, the solutions are going to need to be pretty fundamental, and, as a result, they will also take a long time to implement. So, we need to be in this for the long haul.

The good news, as I see it, is that the 2009 Cassatt Data Center Survey suggests that the magnitude of the problem is starting to be seen and understood. IT and facilities groups should be given credit for the initial actions they seem to be taking in the areas that they have under their control. It's a great start. Now, if you combine these small initial steps with a bit of economic pressure (which the outside world is adding to the mix quite effectively on its own right now), who knows? Maybe even some of these outdated habits will fall by the wayside.

If you'd like a copy of the 2009 Cassatt Data Center Survey results, e-mail me at

No comments: