Designing your data center for the worst case
Have you thought about low load conditions?
On a recent project status call an engineering firm’s mechanical engineer stated, “We always design for the worst case.” The focus of the call was to troubleshoot operational problems associated with an air handler that was experiencing problems due to low load conditions. Weeks of effort and multiple attempts to get the unit to operate had resulted in repeated failures. On the call were the design engineer, the installer, the air handler’s manufacturer’s representative, the commissioning agent, and various other entities. Basically, everyone agreed the unit would work properly if it was seeing higher load conditions. The problem was it was still early in a phased construction project and the existing conditions were far from maximum load. Quite the opposite, the loads were fairly small, so the unit was tripping safeties associated with low load demands.
What struck me as interesting was that in reality the “worst case” scenario was not what would be encountered at rated design loads, but was obviously occurring at the opposite end — at loads associated with “Day 1” conditions. Some of the recommended solutions would have significant negative impacts on the unit and system including customizing what was supposed to be a standardized unit, writing new complex controls programming to implement an equally complex new sequence-of-operations, and abandoning some of the unit’s features and components due to oversizing to try and get the unit to perform “adequately” at low loads. There was no longer any discussion regarding optimization, reliability, or energy efficiency.
The call was collaborative and all participants maintained a spirit of cooperation and willingness to do what was necessary to resolve the unfortunate situation. The owner and his operating staff were supportive with open minds in regard to entertaining “temporary fixes” to get the spaces conditioned and allow the construction to progress. The discussion regarding what would constitute an acceptable “permanent solution” was deferred to another day.
Follow-up emails started moving in a somewhat more defensive manner. The air handler manufacturer took the position that his unit was a high-quality product that was being asked to operate in conditions it wasn’t designed for. The installer reiterated that he installed the unit per the design and was going over and beyond his contractual obligations to get the unit to operate. The design engineer started citing requirements that the unit has to operate over the entire range of normal operations. The owner rightfully took a wait-and-see posture until a final solution could be determined, and the commissioning agent remained neutral, focusing on verifying the installed system performs as required by the design and specifications. The potential for changeorders, litigation, liquidated damages, errors and omissions, and various other situations we all try to avoid grows as each day passes by.
So I asked myself, if the engineer has designed hundreds of HVAC systems, the installer has installed hundreds of HVAC systems, the manufacturer has built hundreds of air-handling units, and the controls technician has programmed hundreds of systems, why are we having these issues? What is so unique about this project? And the answer I came up with is “nothing.” It is just a matter of the project team failing to understand that the “worst case” is not always the rated design conditions. Often it is the other extreme. It’s the low loads associated with Day 1 conditions.
Data Center Conditions
There is perhaps no industry where this is truer than the data center and datacom industry. Most non-data center facilities are built to accommodate load profiles that aren’t anticipated to grow significantly over the life of the building. Office buildings, laboratories, factories, hospitals, and assembly plants are typically built with a well-defined load profile that may periodically ramp up and down, but isn’t expected to grow significantly over the life of the facility. Data centers on the other hand are typically designed for ever growing load profiles. Rated design loads can be a magnitude or greater than those expected on Day 1.
Further compounding this dilemma is the need for redundant infrastructure. Take the case of a Tier-4 facility that specifies a 2(N+1) UPS system. If “N” equals one, then you end up with 2x (1+1), or four UPS modules. So during normal operations you would never see more than 25% rated load on any one module, and that’s at rated load. If the Day 1 loads are only 25% of the rated design load, then each UPS module could be operating at 6.25% of their full capacity. Considering that most data centers have a policy that limits “usable” capacity to 90% or less of the rated capacity, the resulting Day 1 load for the example above would be 5.6% of the full rated load of each module. And this doesn’t take into consideration any “margins-of-error” that the manufacturer includes when rating his equipment or the design engineer includes in his calculations for specifying and selecting equipment.
In data centers, the mechanical systems are designed predominately to cool their electrical counterparts, along with some additional “margin for error.” Whereas most electrical systems can be “turned-down” and operate at very low loads (though at reduced efficiencies), mechanical systems are less accommodating, and it can become physically stressful when you start frequently cycling large equipment on-off. I know of more than one instance where data centers ended up deploying load-banks (resistive heaters) in computer rooms to supplement the small Day 1 IT loads just to produce sufficient heat load to keep the chilled water plant on-line. And of course, the resulting energy efficiency on Day 1 can be horrendous. When you are cycling equipment on and off, operating chilled water and HVAC systems at very small differential temperatures, and operating equipment far from the optimal efficiency bands along with small IT loads, you end up with extremely high PUEs.
So the moral of this story is that advance consideration and planning during the project programming and engineering phase to clearly define the entire realm of operating conditions will result in the best design. Not only should safety margins be included in sizing equipment, systems, and sequences-of-operations to meet maximum loads, but similar safety margins should be included to accommodate smaller than anticipated Day 1 loads. It is my experience that far fewer facilities ever reach their ultimate maximum loads than there are of those that see significantly smaller Day 1 loads than was predicted and programmed for. In some cases, the worst case (and the one which presents the biggest risk to continuous operations) isn’t the unlikely failure scenario and emergency response resulting from operating at high loads. It is the Day 1 low load conditions which require the oversized supporting critical infrastructure to operate outside of their range of normal operations.