Understanding the Metrics for Chilled Air in Data Centers
Perhaps the only thing growing faster than energy use in data centers is new metrics and acronyms to measure and control them. One way to consider the problem of data center energy use is to view the data center as a giant air pump. In a perfect world, every cubic foot of air produced by the CRAC units would find its way to the server inlets.
Unfortunately the world is rarely perfect. This makes air-management metrics a powerful tool for improving the energy efficiency of data centers. By quantifying airflow performance, facility executives can measure and quantify the effectiveness of their data center cooling systems or changes that they make to their cooling systems to increase air cooling efficiency. This approach applies to both traditional raised floor designs and non-raised floor air-conditioning designs for data centers.
What follows is a discussion of a conceptual approach to air management metrics that address bypass flow and balance of server and CRAC unit flows, as well as re-circulation.
Negative pressure flow is another airflow metric, but it generally does not have a significant impact.
A more technical treatment of these metrics, which includes formulas and factors in negative pressure flow, is available at www.facilitiesnet.com/11481bom. These metrics establish how well the cooled air from the CRAC units reaches the server inlets. Ideally, all of the air from the CRAC units would go to the servers. With these metrics, facility managers can create a picture showing how far from the ideal a data center is, and use that picture to improve air management.
For these metrics, the main parameters are air mass flow rates and temperatures, which are used to express the quantity and quality of energy supplied to the servers. These air management metrics are based solely on the consideration of sensible loads. For this conceptual model, the following flow rates are considered.
CRAC flow: Total air-flow rate produced by all operating CRAC units in the data center is normally greater than that required by the servers, due to part load operation of IT and communications equipment, and redundant CRAC units that are kept operational.
Bypass airflow: Air that leaves the floor grills and returns directly to the CRAC unit without cooling servers.
Re-circulation airflow: Air discharged from servers that returns and mixes with chilled air entering the servers.
Using airflow rates and temperatures, it is possible to develop formulas for bypass ratio (BP), recirculation ratio (R) and balance ratio (BAL). Balance shows the ratio of air from the CRAC units that reaches the server inlets. The ideal balance is 1, showing that all the air from the CRAC units is going to the server inlets.
The scale for these ratios is 0 to 1. For bypass and recirculation, the ideal result would be 0 — that is, no bypass air and no recirculation.
On the other hand, if the bypass ratio is 1, all CRAC cooling air would be bypassed, with no air available for servers. And similarly, if recirculation is 1, all server air would be re-circulated, with no cooling air entering from the CRAC units.
Practical Applications
Even if difficult to precisely measure bypass, recirculation and balance, there is always a real specific value that is characteristic of every data center.
In practical terms, the CRAC system should be controlled based on supply air and not on return air. Unfortunately, few data centers are operated that way. Controlling on supply air improves the uniformity in server intake temperatures, avoids unplanned latent cooling, boosts the uniformity of chilled water openings on the chiller water CRAC units and reduces fighting among controls.
While retrieving exact measurements is difficult, the authors obtained representative indices for different types of air management solutions.
Legacy data centers, particularly if not well-managed, tend to have high levels of both bypass and re-circulation flow rates, which are increased when subject to higher density loads.
Cabinet inlet refers to closed cabinets with air entry from below and with fans on top of the cabinet to extract heat. Generally there are no blanking plates within the cabinets, and inlet and exhaust air streams are not segregated, producing high levels of re-circulation. Bypass air is better managed unless there are a large number of floor grills, which would bypass air back to the CRAC units.
Back-to-front arrangements tend to have low bypass but high levels of re-circulation from the back of the cabinets to the front of the cabinets in the same aisle.
Cold-aisle/hot-aisle arrangements, if properly designed and operated, can address the main problems of re-circulation. However, particularly at low loads, there will be high levels of bypass air, unless some type of variable air-flow control is fitted.
Cold-aisle/hot-aisle schemes with all best design practices — such as CRAH units in galleries with high bay ceiling (or return ceiling plenum with ducted CRAHs), blanking panels, cable brushes, foam pillows, and continuous rows — have shown reduced recirculation and bypass levels. However, a big opportunity still exists.
For high density loads, there are several designs whose basic intent is to contain and separate the cold air from the heated return air on the data floor: hot-aisle containment; cold-aisle containment; contained rack supply, room return; room supply, contained rack return; contained rack supply, contained rack return. The separation can be achieved with hard surfaces — "glass" or strip curtains for example. Contained cold/hot aisle is a relatively new approach and performance data indicated very low levels of bypass and recirculation.
Ideally, the data center would have zero bypass and recirculation air flow, and there would also be a balance between the CRAC and server air requirements. But that is far from the case in reality. From surveys and assessments carried out by the authors, on average, half the air flow rate produced by the CRAC units bypasses IT equipment and returns to the CRAC units without doing any useful cooling. And approximately half the air intake to the servers is exhaust air from servers.
