In earlier posts and conversations, I’ve mentioned my interest in container data centers and the opportunities they bring for certain types of implementations and use cases. As a big fan of agility in IT in order to meet the dynamic demands of the Business, containers can be a valuable weapon in the arsenal.
In my conversations with Clients about the applicability of containers to their data center roadmaps, I often find myself involved in philosophical debates.¬† Part of this is due to the newness of the technology as well as preconceived notions of use cases for containers.¬† Part too, is due to the real fact that containers are very strong solutions in some use cases, and more of a strategic option to be evaluated in others.
One could say that containers are at the same time very similar to and very different from conventional approaches to creating data center space.¬† The similarities and the differences are amplified depending on the features of the particular container in question (i.e., which manufacturer’s product) as well as from the intended programming of the data center floor space.¬† In order to consider the full range of options available through the use of containers, we have to consider the options of the (off-the-shelf) container fully and densely populated with servers and/or storage as well as the use of a container just as a (more capable) shell for IT floor space.
Let’s start with the easy one- how are containers similar to traditional data center space?¬† Well, for one thing, they can be viewed as just that- space.¬† This space comes in rather small, modular increments.¬† Current container products come in 20′ and 40′ flavors, so one could view this as say, 300 square foot (SF) increments of IT space.
That similarity carries a corresponding difference though, if you will.¬† Containers have at least some degree of MEP infrastructure on board that supports very high density IT yield of that space.¬† For example, Rackable’s Ice Cube container can pack over 22,000 cores in 1,400U over 28 racks (using Rackable System’s server hardware) in that 300 SF and has cooling capacity to spare.¬† Because of that built-in MEP capability, there are efficiency differences too in the way one would plan the overall IT floor.¬† Space that might be chalked up to inefficiency because of on-floor CRAH, CRAC, PDU’s, and so on would be calculated differently.¬† Because the container vendors out there now are all selling a range of product variations, and because one’s data center architecture plan can vary in its approach to MEP space versus IT space, it’s tough to say what that efficiency calculation ought to look like.¬† Suffice to say though, that it’s a different design approach to the IT floor when containers are involved (‚Äògood question to throw at your data center consultant).¬†
When we focus on that subject of how the data center architecture allocates MEP and IT space, the conversation becomes even more philosophical and subject to lots of “what-if’s.”¬† This is necessary, actually, because (I know, we’ve said it a thousand times already) the MEP infrastructure is by far the dominant driver of data center construction costs.¬† In this context, the spectrum of solution opportunities begins with a full container-based data center design all the way through the use of containers as a use-case specific or a contingency component in an otherwise traditional mission critical facility build.¬† As such, the impact on MEP architecture and MEP planning will be driven by how the container concept plays into the overall data center plan, and what MEP capabilities come with the container to begin with.
If one considers a data center in which the floor space component (needed square footage of IT space) is delivered either through traditional raised floors or containers, we remove many of the advantages facilitated by the container concept.¬† In this approach, the MEP infrastructure for whatever tier level one intends still must be built and whether or not we use containers the overall construction costs are not impacted to a significant degree.¬† Containers do provide modularity and mitigation of risk of overbuilding from a space perspective, but it is difficult to claim a significant cost advantage in that alone (not to mention that the Cost/SF of the IT floor alone is approximately 50% higher if delivered via a container).¬† A possible exception is in the case of containers with fully self contained MEP, but in order to comment on an advantage one way or another we would need more case-specific information.
The strongest advantage of the container solution becomes apparent when one considers a container fully and densely populated with processors and/or storage.¬† This is the true power alley of containers.¬† They provide a very cost effective way to scale very dense computing footprints.¬† The on-board MEP of container solutions are designed to power and cool a very high density of contained IT equipment.¬† Much the same as one would build an enclosure with inline cooling to address a very dense hot spot on the raised floor, the fully populated container is a special instance of that sort of situation.¬†
Of course it’s still the case even with containers that the MEP costs are the dominant up-front investment component.¬† Keep in mind though that a densely populated container solution can bring an opportunity for a much more favorable TCO model.