Recently I spoke to Henrik Leerberg, Product Line Director, Schneider Electric for StruxureWare for Data Centers and I asked him why aren’t people able to reliably increase temperatures in data centers manually to reduce energy costs?
Henrik told me that even at a simplistic level, having people walking up to every CRAC unit and adjusting the set points manually based on the feel of whether it’s hot or cold is really hard to do well. Not least of all because it’s hard to understand or feel the relationship between a cooling unit and the racks that it’s supposed to be cooling.
To help operators do this Schneider Electric now has a new development called Cooling Optimize within the StruxureWare for Data Centers portfolio. This uses what we call an Influence Map for Cooling that understands how every single CRAC unit is influencing each individual rack on the data center floor.
The software learns and understands that when we automatically adjust individual CRAC units to provide air or cold air in the data, exactly which CRAC unit needs to be turned up or down in order to cool down a particular spot in a data center. It also knows that this particular unit may not even be close to the rack in question. This is the challenge of airflow in that it’s hard to predict or understand exactly where the air is moving and heating up or cooling down the data center.
Given this, I asked Henrik if customers have been surprised as to how their air flows are actually delivered or how the air flow sourced for a particular rack might be coming from somewhere other than where they expected it to?
Henrik explained that customers can see in the software application what happens if we turn on an individual CRAC unit and that this could mean temperatures become hotter in a part of the data centre rather than cooler. This can be a big surprise to customers and be completely surprising how your cooling environment is really working because this type of outcome is not one that is intuitively easy to understand.
For customers this means that by using the software they can start to harness energy savings that would have been otherwise out of reach but without threatening the redundancy of their data centers or their ongoing operations.
Henrik agreed and added that this has resulted in customers walking into their data centers and saying “I think it’s too hot in here” but actually it’s not too hot. Maybe it feels hotter because in the past it was over cooled but now it’s at the right temperature to optimise efficiency
Influence maps can’t be created manually. Once the software is installed into a data center, in the first 6 to 12 hours the software learns the system and the data center. It learns by turning on and off each individual CRAC unit and then seeing how this influences the temperature in the room at various places that reflect the entire space. Once this map is built it becomes a base line for the software to use moving forwards.
It’s a baseline because data centers continually change. Customers may install more servers or perhaps use more compute resources and this changes the picture. So then next week we can create a new baseline and we can do that continually from this original starting point and the system will continue to automatically look to optimise the system for a safe balance of temperature versus energy efficiency.