It’s well understood that data centers are large consumers of energy. In fact, in 2013, U.S. data centers consumed 91 billion kilowatt hours of electricity – enough to power all of New York City for two years. By 2020, this energy appetite is expected to grow to 140 billion kilowatt hours, a 53% increase. This consumption will result in a $13 billion energy bill, and emit 100 million tons of carbon dioxide pollution annually.
It is both environmentally responsible and good business to apply strategies to reduce data center energy consumption. Data centers, especially smaller ones, have become the new polluters. It requires 34 U.S. power plants, each generating 500 megawatts, to power today’s data centers. If current growth projections are realized, we will need to add another 17 power plants to meet the energy needs by 2020. According to an article in Area Development magazine, for many data center operators, energy makes up 13 percent or more of total cost of ownership. However, by understanding the data center operations and sequences, as well as implementing Best Practices, reduction of energy consumption by 20 percent or more can be realized. This would save businesses $3.8 billion annually.
Beyond the Best Practices, here are some of the energy saving practices that T5 recently implemented at its T5@Dallas data center.
Mechanical Cooling Energy Reduction Opportunities
Aside from the servers themselves, Mechanical Cooling is the biggest energy consumer in most data centers, and therefore represents the largest opportunity for energy reduction. We have worked with a number of customers to help them reduce their operating expenses by optimizing their mechanical plants.
We recently completed an initiative to optimize the mechanical plant in our T5@Dallas data center. While the inherent design of our Mechanical Plant is energy efficient and reliable, as we operated the plant through seasonal weather changes, we observed some improvement opportunities.
Starting with an ‘inside-out’ approach, we began with the Data Hall. We knew the customer had done a good job of optimizing the floor grate placement and managing airflow to their I.T. load, however as we began looking at the Chilled Water Loop, we noticed a significant variance in the gallons per minute flow rates than we should have seen for a constant heat load. We began trending the Computer Room Control Valve positions and quickly saw that in one Data Hall, the CRAH PID Controllers had never been set up properly by the manufacturer’s technician. The valves were operating from 0-‐100% OPEN. Once we programmed the PID Controls, this section of the Plant began performing as designed. This change can easily be seen for a single CRAH in Figure 1.
Figure 1: CRAH1-‐10 Control Valve Position Before & After PID Control Implementation
Looking at the Mechanical Plant as a whole, given that the flow variances were only causing some excess Chilled Water Pump energy to be consumed (speeding up and slowing down), this really doesn’t represent a large energy savings. However, it indirectly led us to other problems. More about this later.
Another concern that was noticed by both T5 and our Customer was the running of Chillers during periods when we should have been on Plate and Frame Heat Exchanger. According to our Sequence of Operation, our Chilled Water Supply set point is 55°F, so doing some simple arithmetic, subtracting the Heat Exchanger and Cooling Tower approach temperatures, we should be in Full Heat Exchanger mode at an Outdoor Air Wet Bulb of less than or equal to 46°F. This led us to ask our Controls company to perform a recalibration of the I/O sensors throughout the Plant. Upon further investigation, we found that the Cooling Tower Sump Temperature Sensors were not installed in an optimal location. They were on the perimeter of the Cooling Tower Sump instead of at the Condenser Water Inlet. Colder outdoor temperatures caused the Sump Temperature Sensor to read artificially lower than the actual temperature in the Condenser Water Loop. To warm up the Condenser Water, the automated Chiller Manager then slowed down the Cooling Tower Fans, and even opened the Cooling Tower Bypass Valve, which then raised the Condenser Water Loop temperatures. At the Heat Exchanger, the warmer entering Condenser Water then warmed the leaving Chilled Water, thus causing us to breach our 55°F Chilled Water Loop set point. When this happened, the Chiller Manager started a Chiller to ensure the Critical Cooling Load was protected. This, coupled with the Chilled Water Supply flow variances, added to the problem.
The Cooling Tower Sump Temperature Sensor placement became a significant finding. Once we relocated them closer to the Condenser Water inlets, not only did the Heat Exchanger now work as designed, it solved several other headaches. Figure 2 shows the varying Outdoor Air Wet Bulb conditions on April 9/10 2015, and the proper Mechanical Plant response. As Outdoor Air Wet Bulb decreases, the Plant moves from Full Chiller Operation, to Integrated Heat Exchanger + Chiller, to Full Heat Exchanger. We can see the opposite sequence as Outdoor Air Wet Bulb increases.
Figure 2 – Showing Heat Exchanger Taking On Plant Tonnage as Wet Bulb Drops
As mentioned earlier, the variance in Chilled Water Flow rates were also adding to the breaching of the 55°F Chilled Water Loop set point, possibly due to mass flows hitting the Heat Exchanger and it not being able to cool the water back to set point.
Another initiative we introduced to ensure our Critical Facilities Technicians were familiarizing themselves with Plant Sequences of Operations, was the development of our ‘Top Ten Daily Trend Logs.’ These are to be reviewed by every T5FM Technician on every shift. They are on the front page of the BMS and are a snapshot of the Mechanical Plant Operations over the past 24 hours. Outdoor Air conditions (Wet Bulb) is displayed on almost every graph. We use the Trend Logs as metrics to gauge plant efficiency against changing weather conditions, reinforcing our understanding of the Mechanical Plant Sequence of Operation to the ever-changing weather conditions. This also becomes a predictive tool to observe Plant Operation and help minimize energy waste.
Results and Conclusions
We began the initiative in January 2015, during the coldest seasonal months. Using a common metric for Mechanical Plant Efficiency, kW/Ton, we went back and charted some results, shown in Figure 3. What we observed is not a huge negative sloped line (as mentioned earlier, we have efficient Mechanical Plant designs). Instead we saw a narrowing of the variance. The trend line got thinner, the Mechanical Plant was more finely tuned, and the swings in inefficient operations were reduced.
Figure 3 – T5@Dallas Mechanical Plant kW/Ton Efficiency Improvements: Jan-Apr 2015
Having the right team on board helped drive positive results. T5 set the vision, which was driven partly by our own thirst to make our systems better and more energy efficient, but also by the Voice of our Customer. We then collaborated with our Controls and MEP Engineering company to provide the data and affirm our direction. Finally, the onsite T5FM Dallas team was the ‘smart hands and feet,’ and the Customer provided approvals once the risk analysis was completed.
Because the Customer’s Critical Load is increasing, it is difficult to quantify an actual annualized PUE reduction. However, the next cooling season will allow us to compare the kW/Ton data with this past year’s, and from there, we can calculate a kWh (energy) savings in the Mechanical Plant.
Environmental Conscience and Customer Care
Our commitment at T5 is to both customer service and environmental responsibility, and how we handle energy savings addresses both those concerns. In other words, we love counting the tons of Carbon Emissions we avoided ever producing. We work side-by-side with our customers to identify areas where they can reduce the energy component of Total Cost of Ownership, while never compromising reliability.