Your questions answered: Data center electrical efficiency

Designing efficient data centers is a high priority for consulting engineers. Several questions were not addressed at the May 14, 2015, webcast event; the presenters answer them here.

By Bob Schuerger, HP; and Tor Kyaagba, ESD May 18, 2015

Additional questions from the Critical power: Data center electrical efficiency webcast were answered by the presenters after the live event. These responses are from presenters Tor Kyaagba, PE, CEM, CPQ, LEED Green Associate; Associate, Electrical Engineer; Environmental Systems Design Inc., Chicago; and Bob Schuerger, PE; Principal Reliability Analysis Corporate Lead; HP Critical Facilities Services, El Segundo, Calif.

Question: Do you need a control line to calculate transfer switch capacity loads?

Bob Schuerger: The transfer switches are independent of each other. If one primary system fails, that system goes to the reserve. If a second primary system should fail before the first one is repaired, it would also transfer to the reserve, overload, and go down. That is the primary liability of that design.

Tor Kyaagba: I’m adding a bit to Bob’s answer based on my experience. The block redundant designs I’ve worked on include the following provision to prevent overloading the reserve block. When a single primary block load is transferred to the reserve block, an external dry contact is closed to activate the transfer inhibit function for static switches on all other primary blocks. Subsequent failure of a second primary block is limited to that block and prevented from propagating to the reserve.

Question: When are the efficiencies of the IT loads themselves taken into account (i.e. the power use per flop)?

Bob Schuerger: The efficiencies of the IT load is not part of the infrastructure and therefore not related to PUE.

Tor Kyaagba: I agree with Bob here-this doesn’t affect PUE. We might also want to point out that "power use per flop" isn’t really a measure of efficiency per se (more a measure of efficacy or computational performance) since it’s comparing two different quantities. However, anyone who designs or operates enterprise data centers and has input into the IT equipment specification process would definitely want to take advantage of high efficiency power supplies.

Question: When should we expect to see Service 493-2007 annex "O"?

Bob Schuerger: There is not going to be any addition to IEEE Gold Book. All of the IEEE Color Books are becoming the 3000 Series Standards. The Gold Book is now 3006. When the approvals are complete, hopefully in the near future, the U.S. Army Corps of Engineers will have an update to the PREP data and that will be released as a 3006 standard.

Question: Doesn’t Eco mode make a UPS interruptible? Is the transfer time enough that equipment sees it?

Bob Schuerger: The Eco mode is basically operating on the static bypass switch with the UPS inverter operating but not active. It reacts quickly enough that the load does not see a sag.

Question: Where can we find out more about the potential risks of "Eco/ESS" modes, and transfer time?

Tor Kyaagba: There are a lot of great resources available online (white papers, articles) from equipment manufacturers, consulting engineers, and others in the industry. Transfer time will have to be obtained from the equipment manufacturer. In addition to reviewing their product data, I also recommend witnessing product testing at the factory to verify advertised transfer times.

Question: Do the clients not use "Eco/ESS" only because the risks are unknown?

Tor Kyaagba: It isn’t the only reason. But in a lot of instances, having insufficient data to quantify the associated risk makes adopting Eco/ESS mode less likely.

Question: Do you know of cases where these modes (such as Eco/ESS) caused actual problems?

Tor Kyaagba: I haven’t personally encountered sites that dropped the critical load while on Eco/ESS. I think the primary concern here is as it pertains to transfer time coordination with downstream static switches (STSs). If there is a temporary power disturbance, the potential exists for downstream STSs operate before the UPS internal static switch, inadvertently swinging loads between the UPS sources. This nuisance becomes exacerbated if the power quality issues persist.

Question: Is the efficiency curve for premium efficiency transformers substantially different from the curve for TP-1?

Tor Kyaagba: The standards define the target efficiency levels for different transformer ratings, measured at 35% load and 75 C. The difference between standard TP-1 and premium efficiency is quite evident once both tables are compared.

Question: How do you estimate rack power density?

Tor Kyaagba: Rack power density is neither determined by the electrical engineer, nor does it affect the design PUE. That being said, the kW per rack depends on the application, server configuration, and other factors determined by the IT team. A number of co-location designs I’ve worked on are based on supporting up to 10 kW/rack maximum with an average of 5 kW/rack overall. However, this is based on the specific provider’s knowledge of their existing and potential customer base’s requirements.

Question: What is the kVa per rack rule of thumb?

Tor Kyaagba: See above.

Question: Do you have a cost comparison for each level of redundancy?

Bob Schuerger: The cost comparison for each level of redundancy is done at the when there is a real project by a general contractor who has an estimating group.

Question: Why wouldn’t switches and cables be shown on the reliability block diagram?

Bob Schuerger: The switches and cables are in the blocks. The software does a series combination for each block. So all of the items that are in series, such a terminations, cables, and circuit breakers to a specific connection are all together in one block.

Question: Please define "dual cord load."

Bob Schuerger: A "dual cord load" is a piece of IT equipment that has two power supplies in it and can operate on either one. In other words it has a redundant power supply. Both power supplies feed dc power to a common bus in the IT equipment, so that if power is lost to one side, the other power supply automatically goes from providing 50% of the required power to 100% of the required power.

Question: The system where you showed the PUE to be as 1.07 to 1.12: What type of cooling unit is used, because compressors in cooling units are one of the major power consumption equipment?

Tor Kyaagba: That site uses direct evaporative cooling, and 100% outside air passes through a misting system and is supplied to the data halls by perimeter fanwalls. There are no computer room air conditioning (CRAC) units, chillers, or cooling towers. The compressor-less cooling definitely played a big part in PUE reduction.

Question: What, in your opinion, can be done to expand the use of so called Eco-modes because it can offer several percentage points of efficiency?

Tor Kyaagba: Eco-mode might not be the right fit for every design or every site, so care should be taken before advocating for its use. Nevertheless, educating the end-users directly about how it works, the pros and cons, and potential PUE reduction will allow them make better informed operational decisions. Attending factory witness tests and actually seeing it in action might also help change perceptions too.