Server power consumption estimation remains the biggest challenge for data center designers and has actually increased...
in difficulty as services move to the cloud and hosting sites. There is still no absolute answer and no single equation, but the fundamentals remain the same.
A kW-per-rack approach is more useful than an estimation of the dated Watts-per-square-foot metric. But how accurate your estimate is still depends on how much you know about your existing operation, its probable growth or shrinkage trends, and how much power an IT device actually draws.
There are many ways to make and organize server power consumption estimates, but it's helpful to group similar equipment types or functions into capacity units -- just don't get too granular. A capacity unit may be a standalone big iron system or a cabinet full of the same servers. You should group devices that are similar in function type, but that does not necessarily mean the devices are identical.
A sub-10,000 square foot data center shouldn't require more than eight to 12 capacity unit classifications. Larger operations do not automatically require more capacity units. Usually, you just have more cabinets with close to the same configurations. Hardware isn't installed according to capacity units, but you'll have developed realistic, generic requirements you can extrapolate for the entire space.
You also shouldn't overestimate power draws. IT equipment nameplate numbers can be inaccurate. The legally required connected load figure that's on the equipment can give you an overestimated metric. Look for the demand load metric -- the actual wattage a device draws with a given configuration and normal workload. If possible, use hardware manufacturers' online configurators. Even better, use real data from intelligent power distribution units.
Use the server's power supply rating as a last resort. Despite documentation, a 300-Watt power supply can never draw 800 Watts. That goes for dual-corded equipment, too. Two 300-Watt power supplies share the load, so the server still can't draw more than 300 Watts. Dual-corded servers do not draw twice the power of single-corded ones.
Verify server power consumption numbers
When you design a new project, be sure to research industry standards. Cabinets of 3.5 to 5 kW are becoming less common, as more data centers use cabinets that draw 7.5 to 10 kW. High-density cabinets can easily run up to 20 kW and could be 30 kW or more for high-performance computing infrastructure.
The quantities of each rack type depend on your operation, but, in general, enterprise data centers still operate with about 50% low density cabinets, 35% medium density and as much as 15% high density.
Download this form as a guide to estimating your server power consumption in capacity units.
No matter how you estimate server power consumption, double-check your estimates and divide the power reading from your existing uninterruptible power supply (UPS) by your existing cabinet count. Do the same for your projected cabinet count and total estimated IT power load.
If your projected average is more than 1.5 times above or below your existing power consumption average, take a much closer look at the numbers. Figure out if you expect a significant density increase, infrastructure consolidation plans, a need for higher performance computing or a move to cloud architecture. If there's no apparent reason for a large change in power consumption, reexamine your assumptions and remember that very few facilities ever really operate anywhere near the designer's initial load estimate maximums.
Lastly, examine the real availability need for each capacity unit. It's becoming common in larger facilities to provide two or three levels of UPS redundancy, with similar changes in the cooling infrastructure. If a good percentage of equipment doesn't really need Uptime Tier III availability, a more distributed UPS and cooling design -- with differing levels of redundancy -- could save significant cost.