Data Center Cooling Algorithms: Consolidation – Part 217 min read
A set of Excel tools will not make us a certified PE or replace the need for a PE or architectural engineer to design our data center. Or, obviate the need for a mechanical facilities engineer to oversee ongoing operation; however, such tools can provide us with a door through which we can enter more intelligently into both the design and management conversations. While I have heard from a few site engineers who had applied some of these tools to help review and evaluate their assumptions about plans and executions, which was by the way gratifying, my intent has been focused more on enabling management to better articulate requirements (wishes) to designers and builders and to formulate a vision for variables monitored and managed by any systematic mechanical plant management tool. To that end, since last September I have been dribbling out algorithms for planning and managing the various elements of the mechanical plant. As I wrap up this series, for the benefit of the high level browsers, I am just summarizing the key take-aways from each discussion. Last time I covered the first three and today I cover the last five.
- Effective airflow management can make the difference on whether estimated paybacks will or will not justify an investment in free cooling.
- Partial free cooling is not available with parallel water-side economization.
- A tool for estimating parallel water-side economization free cooling hours must be preceded by a decision on how few free cooling hours can be accessed to allow shutting down and restarting the chiller. This decision point then must be accounted for in how the algorithm scans bin data look-up tables. For example, if the data tells me that I will not get five consecutive hours of free cooling on March 22, then I may choose not to risk the hard stop and start on the chiller for my forecast for energy use for that date. Excel makes this operation relatively straightforward. In actual operations management, this activity has traditionally required human intervention, though applying artificial intelligence to data center infrastructure management should relax requirements for 24/7 on-site facilities engineering.
- Historically, chilled water loops required a lower entering water temperature from the economizer heat exchanger than the condenser required. This difference resulted in investing in separate towers for condensers and economizers (bad investment), inserting a bypass loop to allow condenser return water or data center return water to warm tower water (lost free cooling hours), or creating a dead band temperature range between condenser requirement and heat exchanger requirement (also lost free cooling hours). Fortunately, with effective airflow management, resultant higher data center supply air temperatures reduce the problematic gap between condenser and heat exchanger requirements.
- Parallel water-side economization is typically more conducive to retrofitting an existing space than series water-side economization.
- Without good airflow management, an investment in economizer cooling can sometimes stretch standard ROI and payback guidelines, but with good airflow management, investing in economizers is almost always a no-brainer. (Notice a pattern here?)
- Look-up table weather data base needs to include a dew point section for calculating estimated humidification and de-humidification energy costs.
- Since energy use specifications for humidifiers and de-humidifiers are stated in terms of kW per pound or kilogram of water added or removed from cubic foot or cubic meter of air, energy forecast tools for these operations therefore rely on data about how much water is in that volume of air at all the different possible dew points. This data can be calculated live for every reading, though I favor a look-up table for all the possible dew points for simplicity sake.
- Advantages of having mechanical cooling coils integrated into air-side economizers versus deploying standard precision cooling units in the white space in conjunction with economizers include subtracting the fan heat load from the data center total heat load, elimination of unnecessary fan redundancy, and the typically higher energy efficiency of larger fans.
- A condenser ton is rated at more BTUs than a cooling ton because the condenser load also includes the chiller heat generation.
- Everything else being equal, condenser efficiency increases as ambient temperatures decrease.
- Over-sizing a cooling tower increases access to free cooling hours and reduces fan and pump energy during normal condenser operation.
- Lower condenser ΔTs equate to higher chiller efficiency.
- Fan and pump efficiency will increase as chiller efficiency decreases, and vice versa. Chiller efficiency generally contains the bigger bang for the buck.
- By allowing the supply temperature from the condenser to fluctuate based on ambient conditions, we reap the benefits of a low supply to the chiller without having to pay for maintaining a low ΔT through the condenser.
- The surface area of the core between the hot and cold fluid and the flow rate of the fluids will establish the efficiency of the heat transfer and affect the operating costs of upstream and downstream systems.
- Resistance to flow, the temperature difference between the two fluids, and approach temperature all factor into how the heat exchanger affects operating budgets for surrounding systems.
- Where feasible, over-sizing heat exchangers is always a good idea to accentuate operational benefits of lower approach temperatures and fan and pump affinity law economies.
- No change to algorithm for pressure drop at different approach temperatures and resultant CRAH fan energy.
- No change to algorithm for chiller energy at different leaving water temperatures.
- No change to algorithm for calculating cooling energy with water-side economization at higher supply temperatures.
- The trajectory for energy savings at higher temperatures becomes flatter at very high temperatures due to server and CRAH fan speed increases, but the slope does not reverse when economization and/or chilled water are part of the design.
- Increasing supply temperatures with resulting increased airflow demand may eat into redundant cooling capacity and, in extreme situations may exceed the existing plant’s capacity. Generally, the payback for adding capacity to meet this demand is very attractive.
So there you have it. For those readers who are not happily putting on their Excel hats to start building look-up tables and worksheets, these last two pieces of this series on cooling efficiency algorithms provide the key learning points from each set of algorithms. The only other critical ingredient is the monitoring function inside the white space itself and for that I refer my readers to:
- Data Center Temperature Sensor Location
- The Shifting Conversation on Managing Airflow Management: A Mini Case Study
The ASHRAE guidelines on sensor placement and use have not yet incorporated the peculiar values of containment aisles or chimneys and therefore will not be that useful for precisely managing the mechanical plant. Finally, for a preview of the eventual wrap-up of this series, I refer my readers to Case in Point: Sample Applications of Data Center Economizer Algorithms. In this piece I exercised the economizer algorithms to ascertain the relative merits of different types of economizers in different geographic locations. The concluding installment of this series will include the economizer analysis along with the other five sets of algorithms for the same base IT load situation.
Data Center Consultant
Let's keep in touch!
Cooling Capacity Factor (CCF) Reveals Data Center Savings
Learn the importance of calculating your computer room’s CCF by downloading our free Cooling Capacity Factor white paper.