Modular has been seen by many as the solution to today’s datacenter design problems for the following (claimed) reasons:

  • It eliminates expensive, custom, mission critical design
  • Is faster to build
  • Has a lower construction cost and time risk
  • Allows smaller capacity steps to match demand
  • Is more energy efficient
  • Has lower opex

Let’s look at those claims and see how they stack up.

Lego bricks toy modular child
– Thinkstock / Kobyakov

Design options

There are many options to buy a design as part of your build – either a container style ‘all in one’, or pre-designed buildings e.g. Compass, or just buying pre-integrated components e.g. Excool, AST (Schneider).

When it comes to building your modular data center the time to build is dictated by the longest lead time elements which are the same as normal builds. There are many options to offload the build risk wholesale – Digital Realty, Compass, HP etc., so you don’t have to buy a container.

In operation we have seen no intrinsic maintenance cost advantage in modular designs. You can do capacity steps in any design. Part load efficiency depends on more than what you buy.

From a TCO perspective all the basic components are commodity – there is nothing that special about the components in any particular modular product. From our experience many modular offerings carry a premium price.

What does modular do well?

It provides factory economies of scale – cheaper than buying components and sticking them together yourself. You get reliable pre-integrated subsystems. There is also a reduced risk of designing a lemon (it’s already been tested and used at other sites).

Let’s talk about the real TCO of your new data center. Design has been commoditized by Romonet software there is no secret sauce any longer for planning and predicting costs. Build has been commoditized by the market; a lot of people have gotten very good at building data centers. The operation of the data center staffing and maintenance has been commoditized by the Fixed Margin market, e.g. Norland. Finally, the energy opex has been commoditized – Romonet software, and many service providers guarantee that performance.

There have also been fundamental changes in operational efficiency. In the old designs humans had to intervene to ensure cost efficient operation, e.g. turning off pumps, fans etc. In the new designs that are designed to work at any load level, and it is human mistakes which cause the design to exceed target cost, and energy efficiency. It’s not a situation of don’t interfere; just make sure you follow the prescribed methods.

An example of the real TCO

One of the key things that modular providers put forward is an astoundingly low PUE, even from day one – I refer to this as a marketing PUE (mPUE). As an example we will compare a high capital mPUE of 1.1 (i.e. a 1.1 PUE from day 1 even with only 1kW of load) with a pretty normal indirect air design which we hope will achieve a 1.25 PUE once it’s at a decent load and fully occupied. We will allow this to start off at a much worse PUE than 1.25, which we know it will. I’ve also modeled them as being both based in Houston, Texas.

What we see below is the 10 year energy cost and over that entire period we see a $600,000 difference. This is for 1MW of IT capacity filled out over 4 years, and then running flat at full load for the remaining 6 years.

We can then offset the $600K opex reduction against the increased capital cost of the super low PUE modular ‘premium’ build.

The problem is finance people don’t see it that way – they are going to apply a discount rate to those savings in the future. Using a 10% discount rate (which is actually a lot lower than many firms use – 20-30% is common practise) we see a saving of less than $400K over 10 years on 1MW of capacity. That’s unlikely to be a sufficient saving compared to the price premium that you’re prepared to bear on the capital purchase at this point.

Your TCO is fixed by the time you designed your DC

Another thing we have seen is a big shift when you have control of the TCO of your data center. We have seen that the opex (particularly the energy opex) is now just a function of the IT load and of the weather if your design should be configured to operate at all the part load points. Meanwhile, your capital is fixed when you start construction; you’ve chosen what you’re going to build. Your staffing cost is largely fixed and very predictable while your maintenance cost is fixed by the components you’ve chosen. This means that 90% plus of your TCO is fixed by the time you start construction. 

It’s therefore extremely important that you select your technology options carefully before you commit to your design.

To sum up:

  • The high upfront capital cost of a ‘premium’ modular datacenter doesn’t necessarily make financial sense compared to a standard, less efficient build. You are paying a premium to most modular suppliers.
  • Modular does have its place, particularly when using pre-built elements within your design.
  • It is extremely important to model the TCO of your data center design so you can select which elements are worth buying in as modular – by the time you start pouring concrete, 90% of your TCO is now fixed!

In my next blog, I will look at how to match capacity to requirements and so further lower your TCO.