Let’s talk about density

Sparse vs. dense (let’s not call any names)

Unfortunately, as professionals whose main concern is orchestrating the use of space, we cannot help but look at contemporary data centers and see all the waste: wasted space, wasted construction materials, wasted potential.

Here are a few diagrams to illustrate what we mean:

Here is an architectural section of Facebook’s data center in Prineville, Oregon which went live in 2010 (image source)
Step 1: Let’s highlight the actual data center part of the section
Step 2: Now, let’s highlight the racks…
Step 3: …and the silicon, assuming it takes up 75% of the racks (which is probably too much)
Result: The area of hardware (red) vs the area of the outline of the data center space we highlighted in Step 1

As you see, only about 9% of the total area of a cross section of Facebook’s cutting edge data center is occupied by actual working silicon. Don’t get us wrong, we really appreciate Facebook’s data center efforts: initiating the Open Compute Project alone was — and is — a huge deal! We are only using their data center as an illustration of just how ineffective the spatial component of a data center can be, even in case of an industry-leading specimen.

On the other hand, in the case of Facebook, pursuing density was never the goal. If you have a field in Oregon, who cares, how dense or sparse you go? Vice versa, spacing for optimal air flow and cooling are paramount considerations.

But within the paradigms of edge networking and “data urbanization,“ the situation is different. As one of the recent DatacenterDynamics articles has illustrated on the example of an NTT’s data center in Hong Kong, there is both a demand and a benefit in dense computing environments. And as far as that goes, there is plenty of room for improvement.

Let’s take a look at how much space would be occupied by actual servers in one of best-known urban data centers — 375 Pearl St, a.k.a. Intergate Manhattan — if the whole thing were used as a data center. Using Intergate Manhattan’s marketing information, we can perform the following rough calculations:

Obviously, there are a bunch of assumptions leading to a certain amount of error in our calculations, but you get the idea: a minuscule portion of the total bulk of the building is occupied by actual computing equipment; the rest is mostly cavities (hot and cold isles, circulation), supporting systems and the building’s structural components. If we went a step further and stripped the computing equipment down to actual mother boards, the total “useful volume” would probably shrink another two times: what is currently a 30+ story skyscraper could be condensed to a monolith just a few floors high.

Comparative sections in a hypothetical environment

Of course, one might argue that all the space and equipment we so irresponsibly left out is highly valuable and necessary, plus there are codes and regulations, limiting just how dense you can go. And then there is also the question of why would you do this in the first place?

All valid concerns, we agree. But all we are trying to say is that we could do a much-much better job at “packing” data centers, if we wanted to, saving a lot of space and construction materials along the way.

Just a thought.


Did you find this article interesting? We would appreciate if you could press the tiny heart symbol below or share it, if so.

Also, in case you didn’t know, Project Rhizome has recently been featured in two major online data center publications: DatacenterKnowledge and DatacenterDynamics. Click on the names of the publications to be redirected to the articles.

And don’t forget to get / stay in touch with us via our Web-site, Facebook or Twitter!