Greenpeace has focused on many issues of great import over the years. I like whales, don’t like shark finning, and it’s hard to be a huge fan of testing nuclear weapons on South Pacific islands. Much good work has been done and continues to be done. Over the past three to five years, Greenpeace has focused a portion of its resources on data center power consumption. It’s an important area so should get some focus. In truth some of this focus on the datacenter (server side) is misplaced. It’s actually mostly the “other side” (consumer devices) where most of the power is going. Each device draws very little but, in aggregate around the world, that power draw far exceeds that of data centers. Nonetheless, data center power consumption is of interest for both economic and environmental reasons and it makes sense to focus on improving it. Frequent readers know I spend a lot of my time on renewable energy, data center efficiency, and the application of renewable energy to data center applications. It’s both an interesting and important topic as well as being a key point of competition across the large operators.
If I’m excited about renewable energy, green power, and the environment and am deeply invested in learning and innovating in this area, why isn’t the release of a Greenpeace report on the topic exciting? It’s partly the report format and partly the content. I’ll look at each, but let’s start with what I would like to see as top focus areas for this work. If I was to do a report on renewable power applications in the data center, my focus would be, in order of importance:
- Resource Utilization: This is by far the biggest lever in the industry where the carbon cost to produce servers, storage and networking equipment and the carbon impact to power it all is largely wasted. Common server utilization rates average between 10 and 20% across the industry. Turned around, that’s 80% to 90% wastage and there is no topic more important to address around data center environmental impact than utilization. The industry could easily deliver improvements in the 2x to 4x range, and this is where I focus a large part of my day job. The greenest power is that which is not consumed and yet resource utilization is not one of the four main focus areas of the report.
- Energy Efficiency: I have this as the number two largest lever to reduce the environmental impact of data centers. This is another area where huge gains are possible – factors of two have been hit over the last decade and this is another area where I spend a considerable amount of time. Efficiency is important because all power consumption, regardless of the source, has an impact on the environment. So our first commitment, as an industry and as consumers, has to be to only consume that which we need and to always be looking for ways to get the job done by consuming less. Technically, Energy Efficiency and Mitigation is part of the four main topic areas in the Greenpeace report, but the assessment looks pretty distant from the facts. If you get a chance, read through the results on this category and see if they match your take on where data center energy efficiency innovation is under way and which companies are driving the biggest environmental gains.
- Go where the problems are: My approach to all the engineering topics I work on is to start with a cost or resource consumption model and use that understanding of where the problem is to drive what I do. In this case, that would be understanding where IT power is going and target the areas of largest consumption or most grievous waste. This approach can yield pretty good results and it is the technique I’ve used for more than a decade. In the Greenpeace report they focus on data center power consumption exclusively which is neither the biggest consumer of IT power nor is it where the bulk of the coal is going. That’s actually fine and, in fact, that’s where I chose to focus as well because that’s where I have the most influence at this point. Where the report authors again seem to lose their way is to report only on the world’s largest operators. These operators represent considerably less than 5% of the overall consumption and are actually among the most efficient in the world with estimates showing these datacenters are operating nearly 2x more efficient than the industry average. The results from 2011 European Data Center Summit go back 4 years but this data has the large operators at 0.28% of the world’s IT power budget. What that means is the operators covered by the report represent far less than 5% of the problem and are also the most innovative and efficient operators out there. I understand there is an appeal to cover the brands that people most recognize and it’s easier than spending time on a large number of relatively small players, but it does leave the bulk of the problem completely unaddressed. For example, when the first version of the Greenpeace report was released a few years back (How Dirty is Your Data), it was highly critical of Facebook and yet, at the time, they were estimated to have only 100k servers and were clearly one of the most energy efficient operators in the industry. While I was blogging some of the Facebook energy efficiency innovations, it was a mystery to why they were being held up in the Greenpeace report as one of the worst environmental offenders. It just doesn’t follow from the facts. Any approach that doesn’t start with where the majority of the IT power is going isn’t going to be effective at addressing the problem.
- Renewable Energy commitment: This one is well covered in the Greenpeace report and I agree with the focus.
- Leveraged solutions: This is my final focus area and what I’m after here is making a difference that has the maximum leverage. It’s one of the reasons I chose to work at AWS. Small improvements can have very large impact when fully deployed at very high scale. The Greenpeace report does make an effort to use leverage in that they ask the companies included in the report to publicly advocate renewable energy. I would rather see technical solutions than advocacy but there is certainly some leverage in advocacy since large technology companies absolutely do have broad influence both inside their respective industries and beyond. What’s missing is the largest form of leverage that I currently know of in the industry: cloud computing. Cloud computing is an antidote for much of the problem Greenpeace is targeting and yet this is not part of the assessment criteria.
Let’s look at the last issue in a bit more detail since it is not properly covered in the report yet features prominently on my top level list of how to green the world’s data centers. The most obvious cloud computing contribution is on the #1 problem above: server utilization. All companies have to provision hardware resources to their peak usage but only consume the average. This inefficiency is characterized by the peak-to-average ratio. The smaller the peak-to-average ratio, the better it is for the environment, the higher the resource utilization, and the lower the overall costs. This is good for the environment and good for the involved companies economically. When you combine a large number of statistically unrelated resource consumers, the peak-to-average ratio falls dramatically and fast. For example, tax preparation systems unsurprisingly have a real usage spike at tax time, retail is most busy just prior to Christmas, and there is little that can be done to change either. All businesses operate on some cycle and as these mostly non-correlated cycles are combined in a cloud deployment, the peak-to-average ratio falls fast. The same workloads can be supported at a far lower environmental impact.
These cloud computing gains on utilization are easy to understand and fairly obvious but other gains are potentially even more interesting. In the discussion above, we noted that the big operators actually represent less than 5% of the total IT power consumption. They actually can’t make enough of an influence on the outcome by themselves and so we need a broader program. How to get to the majority not included in this group? This is a difficult and somewhat vexing problem – it’s hard to cost effectively change the hundreds of thousands of small data center deployments spread all over the world. Making things worse, these small deployments can be very inefficient, often they are built using old data center designs, some are really just closets in office buildings, and most of the companies that own them don’t have budget to spend on innovative data center ideas, resources to test new solutions, or the ability to modernizing these low-scale deployments.
The best way to address these small deployments that make up the bulk of the problem is to move them to cloud computing. Suddenly the utilization goes up, they are immediately running on very current, high-efficiency technology, it frees up office space in their facility, and they no longer need to overdrive office cooling systems not designed to cool data centers efficiency to cool servers. It’s all upside, it’s addressing the majority of the problem, and yet it’s not featured in the report.
We talked about the format of the report and why my top environmental focus areas are so different from those adopted by Greenpeace. My top areas are: 1) resource utilization, 2) energy efficiency, 3) go where the problems are, 4) renewable energy commitment, and 5) leveraged solutions. The Greenpeace top items are 1) energy transparency, 2) renewable energy commitment & siting policy, 3) energy efficiency & mitigation and 4) renewable energy deployment & advocacy. These four areas selected by Greenpeace seem to miss the mark and they appear to be very difficult to objectively measure. My biggest concern is that great scores can be achieved in the Greenpeace report while doing a poor job on taking care of the environment and, worse, the opposite can happen as well. Companies can be rated poorly while doing great things environmentally. These four focus areas just don’t look like the most effective places to focus if really trying to green the data center industry.
As well as addressing the format of the report, I promised to also look at the content. This one is a short topic but still a very important one. The Greenpeace report is based upon some combination of Greenpeace estimates and vendor self-assessment. Where I know the data through other sources and can independently assess it, there are serious problems in the data Greenpeace is presenting. Greenpeace is aware of many of these inaccuracies but chooses to continue to publish using them. Using data that is known to be incorrect is somewhere between misleading and irresponsible. And, more important, it’s not necessary — approaches look both easier to execute upon and far more effective at lower environmental impact. I argue above that a different set of focus items will deliver more environmental good and deliver it faster.
I’m not excited by the Greenpeace report because, each year, it represents a lost opportunity to have broad impact on IT power consumption and increase the renewable component of this consumption. Each year is a new opportunity to explain the entire problem to the industry, show where the power is going, and advocate the solutions that will yield the broadest impact. Instead we get to read a faulty assessment of the work being done by less than 5% of the industry.