If I tell you my data center has a power usage effectiveness (PUE) rating of 1.2, you’d likely think I was doing very well at addressing energy inefficiencies. But if I told you my PUE was 1.6, you’d probably think I have a lot of room for improvement. In reality, PUE numbers alone hardly tell the story about whether your data center is efficient.
There are many things that will affect the PUE of a data center such as: the equipment specification, the design (example: free cooling vs. no free cooling), the design tier (the amount of resiliency required in the infrastructure), the operation (such as the temperature at which it operates) and the location (the region in which you’re located).
A PUE of 1.2 might be quite good for a Tier 1 data center, one that is based on an “N” design that has little in the way of redundancy. But that 1.6 number might actually be even more impressive if it’s applied to a Tier 4 data center that is designed to a 2 (N+1) specification – meaning a high degree of redundancy.
The reason is quite simple. A Tier 1 data center essentially has one of everything it needs. If the data center is designed for a 500Kw load, it’ll have a single 500Kw UPS (or two 250Kw modules), a single central plant designed to cool a 500Kw load and so on. Such a data center operating at 90% of its rated capacity, assuming it follows all the usual best practices in terms of energy efficiency, can achieve a good PUE rating of 1.2 or so.
Now think about a Tier 4 data center using a 2N design. To handle that same 500Kw load, the Tier 4 data center will need two 500Kw UPSs (or four 250Kw units), along with twice as much CRAC capacity and so forth. And all of these systems will only be able to operate at about 40% to 50% of capacity because if one fails, the backup has to be able to take on the failed unit’s entire load. And if the data center uses a 2 (N+1) design, it’s even less efficient – operating at only about 33% of capacity.
Companies that operate higher-tier data centers do so for a reason, of course – they can’t afford the risk inherent in lower-tier designs. If they have to sacrifice some amount of energy efficiency to achieve their desired risk profile, that’s the cost of doing business.
But there are ways to achieve a sort of happy medium between lowering risk and increasing efficiency when it comes to a data center. One is to use what I call a hybrid tier design, where you design the data center to meet your risk requirements but in a way that might stray from the “by the book” tier design.
To achieve 2N redundancy when it comes to power, for example, your immediate thought may be to have two power feeds from the electric utility substation. Perhaps that extra feed will cost you $4 million to $7 million. Another way to achieve the same objective is to have 2N fuel distribution on your N+1 backup generator system. That’ll give you the resiliency you’re looking for, but at far less cost, maybe $1.5 million. And, if an ice storm or hurricane comes through, you may well be without utility power anyway, so the fuel distribution method may provide you with a better risk profile.
Here’s another example, using simple nominal numbers, that more squarely affects energy efficiency and PUE. Consider a data center with a 500kW critical IT load. With a high resiliency Tier 4 design of 2(N+1) the data center needs three 250kW modules on each side (A and B) for a total installed UPS capacity of 1.5mW (6 times 250k). The 500kW load is normally split between the A and the B sides, hence each 750kW system is loaded to 250kW, or 33%. Going to a 2N UPS design, there is a total of 1mW of UPS capacity installed with two 250kW modules each (500kW) on the A and B sides. Running normally the A and the B UPS systems will support 250kW each, or 50%.
Now take the same 500kW load and apply a 3/2 or Tri-redundant design where you have three individual 250kW UPSs (A, B and C) with independent distribution paths to the load with an A-B, B-C and A-C cording and distribution method. The total UPS capacity installed is 750kW and running normally each 250kW UPS will have a 166kW load, or 66%. If one UPS should fail, the other two UPS units will split the load of the failed UPS. In this example, each design allows the dual corded IT equipment to maintain one cord on UPS power while a UPS system fails or is taken off-line for maintenance.
As you can see, the latter approach takes the UPS load from a low of 33% in the 2(N+1) design up to 66%, a driver in providing greater electrical efficiency. Risk tolerance, change management methods and work windows as well as business rules all play a role in the decision as each design above has a different risk profile. But thinking outside the “Tier Box” may help maximize efficiency while maintaining the appropriate level of risk, acquiring more efficiencies and reducing both capital and operational expenses.
I’m all for energy efficiency in data centers, but we need to be clear on what terms mean and how they’re applied. So next time you hear someone talk about PUE, ask what tier and/or level of redundancy they’re talking about and you’ll get a better idea of how efficient that data center really is. And maybe it’s time we expanded the PUE measure by applying it to specific data center tiers and designs, so we can give operators at each tier a realistic number to shoot for.