Air cooling, containment and airflow modeling

Most of the world's servers are still cooled by air: fans in each server pull cool air in at the front and push warm air out at the back. The facility's job is to deliver enough cool air to every server inlet and to carry the warm exhaust back to the cooling equipment without the two mixing. That sounds simple, but in an uncontrolled room cold air short-circuits back to the cooling units without passing through any server, and hot exhaust curls over and around racks into server inlets. Managing airflow is the cheapest and most effective way to improve both the reliability and the efficiency of an air-cooled data hall.

This lesson covers air-cooling equipment, the hot-aisle/cold-aisle layout, bypass and recirculation, containment, the fan laws that make airflow management pay, metrics for checking that every inlet is within limits, and computational fluid dynamics (CFD) modeling.

Air-cooling equipment

The units that cool the air in a data hall come in several forms:

• Computer room air handlers (CRAH): a fan and a chilled-water coil in a cabinet. The heat goes to the chilled-water plant (Lesson 5). CRAHs are the most common units in large facilities.
• Computer room air conditioners (CRAC): a self-contained refrigeration unit with its own compressor, rejecting heat to an outdoor condenser or a water loop. Common in smaller rooms.
• Fan walls: banks of many fans and large coils built into a wall of the hall or an adjacent gallery, supplying air horizontally at low velocity. Widely used in hyperscale facilities built on a concrete slab without a raised floor.
• Close-coupled units: in-row coolers placed between racks, and rear-door heat exchangers (coils mounted on the back door of a rack) that cool the exhaust before it enters the room. They handle higher densities, typically 20 to 50 kW per rack for active rear doors, by shortening the air path.

Air is delivered either from a raised floor, a pressurized plenum (typically 600 to 1000 mm deep) from which it rises through perforated tiles in front of the racks, or directly into the room or aisle from fan walls or overhead ducts. Raised floors were standard for decades; many new halls use slab floors, which carry heavier racks, cost less and avoid the problem of cables obstructing the plenum.

Hot aisles and cold aisles

Because servers breathe front to back, racks are arranged in rows facing each other across alternating aisles: fronts face fronts across a cold aisle, where cool air is supplied, and backs face backs across a hot aisle, which collects exhaust for return to the cooling units. This simple arrangement, standard since the 1990s, keeps exhaust from blowing directly into the next row's inlets.

Even with hot and cold aisles, two flows spoil the separation:

• Bypass: cool supply air that returns to the cooling units without passing through IT equipment, through misplaced perforated tiles, cable openings in the floor, gaps between racks or empty rack spaces. Bypass wastes fan energy and lowers the return temperature, which reduces the cooling units' capacity and efficiency.
• Recirculation: hot exhaust that finds its way back into server inlets, over the tops of racks, around the ends of rows or through empty rack spaces. It raises inlet temperatures, typically at the tops of racks and the ends of rows, and operators often respond by lowering the supply temperature for the whole room, which wastes energy.

Basic airflow management

Much bypass and recirculation can be eliminated cheaply:

• Install blanking panels in every empty rack unit, so exhaust cannot flow forward through the rack.
• Seal cable openings in raised floors with brush grommets, and seal gaps under and between racks.
• Place perforated tiles only in cold aisles, matched in number and open area to the load of the racks in front of them; remove tiles from hot aisles and open areas.
• Keep rows continuous, without gaps, and fill the ends of rows.
• Return air from high in the room or from a ceiling plenum above the hot aisles, so the hottest air is captured.

Containment

Containment physically separates hot and cold air with panels, doors and roofs:

• Cold-aisle containment (CAC): the cold aisle is enclosed with doors at the ends and a roof over the aisle. The rest of the room becomes a hot return space. It is easy to retrofit in raised-floor rooms, but the room is hot, which is uncomfortable for staff and affects any equipment outside the enclosures.
• Hot-aisle containment (HAC): the hot aisle is enclosed and its exhaust ducted to the cooling units, often through a ceiling plenum. The room stays at the cool supply temperature, and a large volume of cool air buffers short interruptions in cooling. It is common in new builds.
• Chimney (ducted exhaust) cabinets: each rack has a duct from its rear to a ceiling return plenum, a form of hot-aisle containment at the rack level.

With good containment, the cooling units can supply air at close to the required server inlet temperature (instead of several degrees colder to make up for mixing), deliver only slightly more air than the IT equipment needs (an airflow ratio of about 1.05 to 1.2, compared with 1.5 to 2.5 in an uncontained room), and return air at the full server exhaust temperature. All three improve efficiency.

Containment must be coordinated with fire protection: sprinkler heads must be able to reach inside the enclosures (or be installed inside them), and roof panels are commonly designed to drop or shrink away when heated. Fire codes and the authority having jurisdiction should be consulted early. Containment also needs pressure control: if the cooling units supply less air than the servers draw, the contained cold aisle goes to negative pressure and hot air is pulled in through every gap; slight positive pressure in the cold aisle (a few pascals) is the target.

The fan laws

Fan energy is where airflow management pays most. For a given fan and system, the fan affinity laws relate flow, pressure and power to speed \(N\):

Because power varies with the cube of speed, a fan delivering 80% of its flow uses only about half its full power. Electronically commutated (EC) fans and variable-frequency drives make continuous speed control practical, and modern cooling units adjust fan speed to hold a set pressure difference across containment or a set server inlet temperature.

Raised-floor airflow limits

A perforated floor tile with 25% open area typically passes about 0.2 to 0.3 m³/s (roughly 400 to 600 CFM) at normal underfloor pressures; high-flow grates with about 50 to 60% open area pass two to three times as much. These numbers limit the density a raised floor can serve.

Is every inlet within limits?

Because the ASHRAE limits apply to each server inlet, airflow is assessed by measuring (or modeling) inlet temperatures, typically at the top, middle and bottom of each rack front. The Rack Cooling Index (RCI), developed by Magnus Herrlin and widely used in standards and CFD reports, summarizes how many inlets exceed the recommended maximum and by how much:

where \(T_x\) are the inlet temperatures, \(n\) the number of inlets, \(T_{\max,\text{rec}}\) the recommended maximum (27 °C) and \(T_{\max,\text{all}}\) the allowable maximum (for example 32 °C for class A1). An RCI of 100% means no inlet exceeds the recommended limit; 96% or more is generally considered good and below 91% poor. A matching RCI_LO measures over-cooling below the recommended minimum. The Return Temperature Index (RTI), the ratio of the cooling-unit temperature difference to the IT temperature difference, indicates bypass (below 100%) or recirculation (above 100%).

Computational fluid dynamics

Computational fluid dynamics (CFD) solves the equations of fluid flow and heat transfer numerically on a three-dimensional mesh of the hall, predicting air velocities, pressures and temperatures everywhere, including at every rack inlet. Data center CFD tools model the room, raised floor, obstructions, perforated tiles, cooling units and the airflow and heat of each rack. They are used:

• In design, to compare layouts, containment strategies and cooling unit placement, and to check that the design meets RCI targets at full load.
• In failure analysis, to show what happens when a cooling unit fails or is under maintenance: which racks overheat and how quickly.
• In operation, as a "digital twin" calibrated against measured temperatures, to decide where new equipment can go without creating hot spots.

A CFD model is only as good as its inputs. Rack airflows are often estimated from nameplate power rather than measured, cable obstructions under the floor are hard to represent, and small leakage paths matter. Models should be calibrated with measurements before their predictions are trusted for decisions.

Lesson 4 looks at the equipment that cools the air: how coils work, perimeter units, fan walls, in-row coolers and rear-door heat exchangers, and how to size them.