Capturing the essence of in-row cooling
Tue 11 Aug 2015 | Victor Avelar
Data centre operators and designers need to be aware of how all aspects of the cooling infrastructure work so that they can design their cooling infrastructure efficiently and respond optimally to any changes.
In-row cooling, also called close-coupled cooling or row-oriented architecture, has been a familiar concept in data-centre cooling for about a decade. However, some misconceptions about just how it works can lead to errors in its application. A common error is to think of row-based cooling as a “cold-air supply” architecture instead of a “hot-air capture” architecture, which can cause unnecessary or counter productive steps to be taken in data centre cooling operations. Used properly, in-row cooling can be efficient and effective; if there are misunderstandings about its function, the results may be disappointing.
The goal of any data-centre cooling system is to remove the heat added to the air by the IT equipment. A metric for measuring the effectiveness of how hot IT exhaust air is captured by cooling equipment (or how cold air is supplied to IT inlets) is called the capture index. This is the fraction of hot air exhausted by IT equipment that is captured directly by the row coolers within the same pod. Capture index is based solely on the airflow patterns associated with the supply of cold air to, or the removal of hot air from a rack and is typically a rack-by-rack metric with values ranging between zero and 100%; higher values generally imply good cooling performance.
The hot-air capture index is a useful tool for optimising the placement of racks and cooling units in a pod, where a pod is a pair of rows of cabinets of IT equipment arranged so that hot exhaust air is directed to the common aisle between them, the so-called hot aisle. The design goal is to ensure that all exhaust air is captured by the cooling units so that there is no net heating of the room. Typically, the hot air capture index target for each rack should be 90% or more, which can be achieved by deploying a number of techniques at the rack level, including sealing off unused positions to prevent hot air escaping the hot aisle and placing row coolers in the most effective positions.
Row coolers capture hot IT exhaust air and neutralise it before it has a chance to mix with the surrounding air in the room or circulate to the front of an IT rack. The three key design attributes are back-to-front airflow, a rack-based footprint, and variable cooling-capacity devices.
Row coolers are designed with back-to-front airflow with the rear of the cooler facing the hot aisle and small fans distributed across the height of the cooler to collect the hot exhaust air from IT equipment. As the fans are distributed across the height of the cooler, they are able to collect the air directly and uniformly, thereby preventing hot spots from occurring.
The hot air capture index is highest for IT racks closest to the cooler, therefore coolers are designed with a footprint similar to an IT rack, coming in either full-width or half-width formats. This rack-based footprint allows coolers to be distributed easily throughout a pod of IT. Note that when a pod is fully contained (including aisle doors and ceiling) a high hot air capture index is ensured regardless of the cooler’s position in the pod.
Row coolers are also designed with electrically controlled fans, whose speed can be adjusted according to IT load by sensing the inlet temperature of nearby racks or the IT room.
Some row cooling units feature turning vanes which are used to direct cold air into adjacent racks. This shows a misunderstanding of how the concept of hot-air capture works. If the design attributes mentioned above are properly incorporated into the design of a cooling system there is no need for turning vanes.
If all of the hot exhaust air from IT equipment is captured and cooled before it has a chance to mix with the surrounding air in the room, the remaining space in the room maintains a cool ambient temperature. Therefore it doesn’t matter where the cold air from the row cooler goes; it only matters that the hot air from the IT equipment has been captured and neutralised.
The use of turning vanes, as well as being an unnecessary additional capital cost, introduces airflow issues that further complicate the cooling process. Typically the airflow leaving the turning vanes is at a higher velocity and perpendicular to the airflow in the adjacent IT rack. The high-velocity airflow creates regions of low pressure in front of the rack and therefore interferes with the airflow. This causes highly variable conditions from rack to rack, making the monitoring and management of cooling more difficult. Turning vanes also increase the cooling unit fan power to overcome the resistance (i.e. pressure drop).
In all, the use of turning vanes increases both capital and running costs for no appreciable benefit.
Another misconception is that row coolers are needed in every row for optimal cooling. However, if a high hot-air capture index is achieved, it doesn’t matter if all row coolers are on one side of a two-row pod. It is possible for row coolers in a single row to cool racks of both rows.
A third misconception is that row coolers can’t cool loads outside their pod. This implies that row coolers are only for spot cooling pods, or cannot contribute to the cooling of the room overall or cannot cool standalone pieces of equipment such as storage units on the periphery of a data centre.
Again, a full understanding of how row coolers capture hot air rather than distribute cold air refutes these misconceptions. Indeed, the most predictable way to cool standalone ancillary equipment is to place a cooler next to it. But if there is no cooler next to such equipment, and it is in a computer room with, say, a two-row pod of racks that do contain row-cooling equipment, then the effect of the cooling equipment inside the pod can help to cool the standalone ancillary equipment.
It works in this manner. As row coolers are variable capacity devices they are capable of over-supplying cold air to the cold aisle. When the system is in balance, the supply temperature from the cooler is equal to the IT inlet temperature. If an extra piece of equipment is added, outside of the pod, the overall temperature of the room increases due to air mixing.
But then the row-based cooling units sense this increase and respond by increasing cooling capacity thereby neutralising the hot air. This increased cooling capacity is achieved by increased air flow (row coolers increase fan speed) and or lower supply air temperature (row cooler increase chilled water valve flow). These adjustments occur initially but remain unchanged once the data centre reaches steady state. In this way, the in-row coolers inside the pod can assist in the cooling of standalone equipment outside it.
Understanding the principles of in-row cooling can help to deploy cost-effective efficient cooling strategies and avoid the expense and additional headaches caused by the addition of unnecessary and ill-advised cooling methods.
For more information on this topic, read Schneider Electric white paper 208