I recently participated in a discussion on LinkedIn about the best, most energy-efficient way to control temperature: either the classic sensor in the return of each individual CRAC/CRAH or supply air either by under-floor sensors or sensors in the cold aisles. Of course, that just led to the inevitable discussion of where and how many sensors would be required, and how would the control system decide what to do if there were some areas that were too warm and some that were too cold.

I wrote that I believed that the IT equipment is already aware of the intake temperature and, that ideally, sensors placed everywhere would ultimately become unnecessary, once DCIM systems become truly intelligent gateways between facilities and IT.

To some it would seem to be considered an “unnatural act” if the communications and management IT protocols of TCPIP, SNMP, and IPMI were to be utilized, interchanged, and acted upon by BMS facilities control systems, which are primarily based on BACNet and Modbus. 

A few DCIM vendors are beginning to include this capability. Some of the early work includes a proof-of-concept project done at Lawrence Berkeley National Laboratory in 2009 (btech.lbl.gov/papers/3137.pdf), which was cited by Bill Tschudi of LBNL in the
LinkedIn discussions. Unfortunately, this is still not the mainstream direction for solving the cooling control system conundrum. 

ASHRAE’s 2011 Expanded Thermal Guideline was prepared with the participation of the major IT equipment manufacturers, who endorsed and supported the higher allowable temperature ranges that make some data center designers and operators a bit uncomfortable, such as the class A2 maximum of 95°F, which is what almost every server’s specification has been for the past few years. Yet, if the warmest temperatures were fully implemented in order to save energy, it would imply the need for a more sophisticated method of temperature control than simply sensing return air at the CRAC.

However, the main reason that this methodology is not readily accepted is the perceived danger that if facilities personnel and equipment were to directly interchange information with IT personnel and IT equipment, it would cause a massive break in the space-time continuum, possibly causing data center cooling control systems to shift forward in time by 20 years, or as some facilities naysayers fear, completely shutdown the cooling every time IT patches a server.

I have always liked the idea of controlling a cooling system based on information derived directly from IT equipment, rather than by installing three sensors per high-density racks and “hoping” that they will provide enough information to control the cooling system as a whole. As many in the discussion pointed out, the sensors do not really provide any localized control over different racks in different areas with a dynamic heat loads cause by a combination of visualization.

The “unintended impact of Energy Star servers,” as I like to call it, is the result of widely varying heat and airflow requirements (many have a 1 to 5 min-max ratio) of the new Energy Star-rated servers. This ratio causes chaos with airflow design assumptions and actual operating conditions and produces “traveling” hot-spots, as well as over- and under-cooling as heat loads change dynamically with computing loads

Nonetheless, assuming that we could convince facilities and IT to cooperate, we should also look beyond just obtaining temperature information from the IT gear. The 800-lb gorilla in the data center is managing and balancing airflow. In our ideal world, the airflow though the cooling system would perfectly match the CFM of the IT gear, even as the CFM requirements change dynamically—et voilà, no bypass air, no recirculation—a perfectly optimized airflow balance.

However, the dynamic nature of server heat loads also means that the fan rates have a very wide CFM (and, of course, power) range. The older servers had simpler fan control algorithms. The newest have very sophisticated fan controls systems that can also presumably report (and perhaps request) that the fan in the CRAC/CRAH match the CFM requirements of the IT equipment.

While this control can occur in very tightly contained systems, such a rack-level containment system with a directly attached cooling unit, it is not really possible with centralized cooling with a common open airflow.

As I was writing this column in mid-February, HP announced new G8 servers, which seem to have a highly developed airflow control system that they claim will reduce fan energy (http://www.missioncriticalmagazine.com/blogs/14/post/84765-hot-aisle-insight--new-hp-g8-servers-staying-cool---using-less-air).  If proven successful, I am sure that other manufacturers will follow shortly.

But will we then take the quantum leap and begin to design and control our cooling systems (temperature and airflow) based on IT equipment requests?  Will we move from the traditional facilities “call for cooling” request, coming from the simple return air temperature sensors or bravely forge ahead and utilize information from the internal sensors of the IT equipment—which can provide not just air intake temperature, but also real-time heat load and airflow requirements. Then we will have a truly energy efficient, “smart data center.”

And, if that really does happen, will pigs begin to fly? 


Reprints of this articleare available by contacting Jill DeVries at devriesj@bnpmedia.com or at  248-244-1726.