Data Center Temperatures and the News

I use a variety of ways to track the news these days and few use traditional paper publications.  I mostly use web-based and, yes, cloud computing applications to manage the inflow of news. These primarily consist of Google Reader, Google alerts, Twitter, and email newsletters. I bring this up for two reasons:

#1. Because unlike physical magazines or journals I don’t actually own any of this content I read.  It’s all stored somewhere else, in various data centers around the world, waiting for me to access it.

#2. It’s a great example of how web-based tools can increase the efficiency of my life, and even decrease my energy footprint.

This latter point is important, because it’s what a lot of us people interested in data center efficiency stress all the time: we don’t want people to stop using data centers and the tools they provide, we just want data centers to follow best practices and minimize their energy footprint while providing the same level of service.

As an example of my news review practices, I just skimmed the titles of probably over a hundred articles in a pretty short time (~30 minutes). Of the few articles I actually skimmed or read, I found this interesting blog post on the data center temperature debate. This is not a blog I follow, or an issue I follow particularly closely, but after scouring my many news sources this is the article that caught my eye enough to post about.  It does a good job of explaining the issues around data center temperature setpoints and the arguments of whether they result in energy savings.  The bottom line, as usual, is that energy savings will likely depend on your specific situation – if it allows you to cool more with outside air it’ll likely help, if it just causes your server fans to kick into higher gear it might not.

I’m not going to take one side or the other on this issue, but what I think is interesting is this: for years, engineers have overdesigned everything. Structures are built with a factor of safety of many times the needed strength; capacities are built out well over requirements just in case more is needed later or something goes wrong; and, yes, data centers are kept much colder than they need to be just to be safe. In some cases these decisions are based more on superstition than sound engineering. One of the things computers and servers have given us is the ability to dial in and optimize some of these design points through computer aided design, advanced monitoring, automated controls, etc.

Here’s where it all comes together:  these new tools should allow a savvy data center operator to dial in the correct cooling level across the data center and not have to significantly overdesign the cooling system or setpoints.  Good monitoring should identify problems quickly or even before they occur.  Good controls should alter settings to compensate for potential problems.  The two together should be tuned to optimize efficiency and decrease costs.  I’d be willing to bet that for the majority of data centers there can be significant energy savings by increasing temperatures, but you need to use these tools to address potential hotspots and to determine if that increase should be 1 degree, 2 degrees or 10 degrees.  Too much or too little and there may not be savings, or worse, there could be an increase in energy use.

So, in summary, we continue to keep using computers and data centers to increase efficiencies across all aspects of the economy, and, of course, use these techniques on data centers to slow the energy growth of these tools themselves.

Now, if I could just get the Economist to forward my magazine subscription to my new address.  How could this possibly take 3 weeks in this day and age? Maybe that will be the subject of another post. No matter, I’ll just read it online…

How big can data centers be? How about 19 football fields?

Today I realized that my understanding of how large data centers can get was significantly understated.  This realization came as I reviewed Data Center Knowledge’s special report on the world’s largest data centers.  I have previously used #5, Microsoft’s Chicago data center, as an example of one of the largest, but was shocked to realize that the largest is almost 60% bigger.  Another interesting result is that seven of the top ten are colocation facilities.  This is significant because it is often difficult to get colocation facilities to engage in energy efficiency programs, especially after they’re operational. The other three – and the only corporate data centers in the top 10 – are all Microsoft facilities. A number of the facilities (including the largest) are also buildings converted to data centers from other uses.  Since these are not purpose-built data centers, my guess would be that they are probably not ideally designed in terms of efficiency.

I’m also disappointed to see that relatively few have energy use or even power capacity listed.  In an environment where power is starting to dominate as the primary constraint on data center growth, wouldn’t it make sense to track a list of the largest data centers in terms of energy use?

Here are some other highlights from the report:

#10. SuperNAP (407,000 SF) – Number ten is notable mostly for its power consumption.  At 250 MW capacity it boasts densities of up to 1,500 w/SF, made possible through advanced cooling using “a high-density T-SCIF (Thermal Separate Compartment in Facility) containment system to fully separate the hot and cold aisles.”

#7. i/o Data Centers Phoenix ONE (538,000 SF) – This one just seems to keep popping up, with “enormous rooftop array of solar panels that will eventually generate as much as 4.5 megawatts of power for the data center, and a thermal storage system that will allow i/o Data Centers to run chillers for its cooling systems at night when power rates are lower.”

#6. Microsoft’s Dublin Data Center (550,000 SF) – This one operates 100% of the time on outside air through the use of economizers and “Microsoft says it can run its server rooms at temperatures of up to 95 degrees F (35 degrees Celsius),” which should give it an efficiency advantage.

#5. Microsoft Chicago Data Center (700,000 SF) – A large portion of this data center consists of double-stacked 40-foot shipping containers that are each filled with up to 2,000 servers.  Containers make the system highly scalable and efficient.

#1. Digital Realty Trust Lakeside Technology Center (1.1 M SF, 100+ MW of power) – In Chicago, this data center used to house the printing presses for the Yellow Book and the Sears Catalog. It was converted to telecom use in 1999 and is now 2^nd largest power customer for Commonwealth Edison.

Some people might wonder why a bohemoth such as Google doesn’t show up on this list? Well, it seems that Google likes to focus on many data centers together on a campus, while Microsoft tends to go big, and the report only looks at individual buildings not campuses.

So how big are these?  Let’s put it in perspective:

1.1  million square feet is equivalent to just over 19 football fields

250 MW is equivalent to the average power use of about 200,000 American homes

These numbers really speak to the massive amount of computing needed in modern society, but this is actually not where the majority of energy use from data centers come from.  According to the US EPA’s 2007 report to Congress, only 38% of data center energy use in the US comes from “enterprise-class” data centers of greater than 5,000 SF.  The remaining 62% is used in the smaller data centers, which means that these smaller data centers offer the largest overall chance for energy savings in this industry.