TRP: Can you take us through how you deal with factors like Uninterruptible Power Supply and redundancy with a modular setup?
PK: We have a centralised UPS system that makes it easier for us to route the power requirement where it's needed. The decentralised UPS that featured in our original design meant that if you had a failure, because you had so many modules, you only impacted one part of of the module's rows. On a centralised system, if you have a critical system failure, you impact half of every row.
TRP: The data centre has a design PUE (power usage effectiveness) of 1.21. How much importance do you place on PUE, and do you see the metric as the dominant one in the industry for some time to come?
MG: I see PUE becoming slightly more sophisticated in time, but I think that it will remain as the headline metric because it's such an easy one to understand. It's also comparable across the industry, so if you're going into a data centre that boasts a PUE, you can compare it against another - so long as you know how to measure it.
Most importantly, within a specific data centre, PUE can act as an overall indicator of whether you're getting more efficient. Essentially, as a data centre gets more utliised, you can track it.
Some of our halls here are down to 1.1 PUE, which is just about as good as you can get. The design PUE for the modules out there is about 1.21, which means that by the time you get to something like 60% full, it will hit that 1.21, and that's industry leading.
There are other indicators, and we use them, but they're sub-level at the moment. Over time they may get more prevalent.
TRP: Carrying on the utilisation theme, how does the old data centre adage "the more energy you use the more you save" apply to a modular data centre?
GM: Utilisation is important, no question. If you're running an empty hall, then the PUE can be very poor. Actually, even in that case, the ones here are very efficient, funnily enough, as the modularity allows you to decide which bits you want to run.
Putting that aside, yes - as soon as you get to over 50% full you can start really getting the benefits of efficiency. The quicker you can get a data centre to that level or above, the better the efficiency you get out of it, so that's what we're trying to do.
The advantage of a modular system is that you can match supply with demand. You will always be, as a data centre overall, in that 75-85% capacity range. When you get to that level you build another module until the whole thing gets full, and you might get to 95%. You'll never get above that as there's always some wasted space.
KF: Can you talk us through the data centre's cooling system?
MG: At the heart of the modular system that Colt has designed is the cooling unit, called the downflow unit, which essentially has three modes of operation. The predominant mode, which allows us to have a very efficient data centre, is to use free air cooling. That's really taking air from the outside, cleaning it and pushing it through the data hall using the cool down equipment. That allows it to be an incredibly efficient way of using cooling.
Of course, when it gets incredibly hot which is about six days a year, you can switch to DX cooling, which is done automatically. There's also an intermediate level of indirect cooling under certain conditions where the humidity reaches different points. That allows you to recycle air inside the data hall but use the cooling from the free air that comes outside.
All in all, the unit itself is incredibly sophisticated, and even moreso going forward because we have the ability to have updatability within the hall. So, you may start off with a hall that has a certain power density that it can look after, but then by slotting in different cooling units on the outside of the hall you can increase the power density.
