Air Cooling Fundamentals

Introduction Picture your data center as a massive computer cluster-thousands of servers processing requests, each one generating heat like a miniaturked CPU under load.

A single rack can generate 15-20 kW of heat, roughly equivalent to ten household space heaters running simultaneously.

Without proper cooling, temperatures would spike past safe operating limits within minutes, triggering thermal shutdowns and turning your infrastructure into an expensive brick.

Air cooling systems are the load balancers of the physical world.

Just as a load balancer distributes network traffic across servers to prevent any single node from overheating, air cooling distributes cold air across your facility to prevent thermal hotspots.

The equipment that makes this possible-CRAC and CRAH units-are your primary cooling infrastructure, working alongside airflow management strategies to maintain optimal operating temperatures between 64-81°F (18-27°C) as recommended by ASHRAE standards.

This lesson arms you with the fundamentals of air cooling design: understanding the difference between CRAC and CRAH systems, implementing hot aisle/cold aisle containment to improve efficiency, and calculating cooling capacity so you can properly size your infrastructure.

These skills are essential whether you're managing a single server room or planning a multi-megawatt facility.

Understanding CRAC and CRAH: Two Approaches to Air Cooling

CRAC Units: Computer Room Air Conditioners Computer Room Air Conditioners work exactly like the air conditioning system in your car or home, but scaled up dramatically.

Each CRAC unit contains its own refrigeration cycle with compressors, condensers, and evaporators.

Think of them as self-contained cooling appliances-plug them in, connect the chilled air distribution, and they independently handle the cooling load.

CRAC units typically deliver 10-30 tons of cooling capacity per unit, with one ton equaling 12,000 BTU/hr or approximately 3.5 kW.

A facility like Digital Realty's Chicago data center might deploy 20+ CRAC units in an N+2 configuration, ensuring redundancy if multiple units fail during peak summer loads.

The downside? CRACs consume significant power to run their compressors.

They're also sensitive to outdoor temperature-when ambient temperatures climb above 95°F, their efficiency drops substantially because the refrigeration cycle must work harder to reject heat.

This is similar to how your application performance degrades when system resources max out.

CRAH Units: Computer Room Air Handlers Computer Room Air Handlers take a different architectural approach.

Instead of generating cold air internally, CRAHs use chilled water from a central plant-think of it as a microservices architecture where cooling generation is separated from distribution.

The CRAH unit contains fans and heat exchange coils but no compressors.

Chilled water (typically 42-48°F) flows through these coils, and fans blow warm return air across them.

Equinix's SV1 facility in Silicon Valley relies heavily on CRAH technology, with each unit handling 50-100 tons of cooling capacity-significantly more than CRAC units.

The central chiller plant serves multiple CRAH units across the facility, similar to how a centralized authentication service handles requests from multiple application servers.

CRAHs offer better scalability and efficiency, especially in larger deployments.

When CyrusOne builds a new facility exceeding 10 MW of IT load, they almost always choose CRAH systems because:

• Central chillers can leverage free cooling (using outdoor air when temperatures permit)
• No refrigerant piping is required throughout the data hall
• Maintenance is centralized at the chiller plant rather than distributed across dozens of units

  Feature	CRAC	CRAH
  Cooling Capacity	10-30 tons per unit	50-100+ tons per unit
  Refrigeration	Self-contained (DX cycle)	Central chilled water plant
  Typical Deployment	Smaller facilities (<2 MW)	Larger facilities (>5 MW)
  Energy Efficiency (PUE impact)	1.6-2.0	1.3-1.6
  Capital Cost	Lower initial investment	Higher upfront, lower operating cost
  Redundancy Model	More units needed (N+1, N+2)	Fewer units with central plant redundancy
  Response Time	Faster (local control)	Slightly slower (depends on water loop)

Hot Aisle/Cold Aisle Containment: Organizing Airflow Like Network Segmentation Imagine if your network allowed any packet to travel any path without routing rules-chaos.

The same principle applies to airflow.

Without proper organization, cold supply air mixes with hot exhaust air, creating inefficiencies that waste cooling capacity.

The Basic Architecture Hot aisle/cold aisle containment organizes server racks in alternating rows.

All equipment faces the same direction, with front (intake) sides facing cold aisles and rear (exhaust) sides facing hot aisles.

CRAC or CRAH units deliver cold air to the cold aisles, servers pull it through for cooling, and exhaust it into hot aisles.

Return air from hot aisles flows back to the cooling units.

Think of cold aisles as your ingress traffic and hot aisles as egress-maintaining separation ensures efficient routing.

Containment Strategies Cold Aisle Containment (CAC) encloses the cold aisles with doors and ceiling panels, creating a pressurized cold air plenum.

AWS uses this approach in several of their us-east-1 availability zones.

Cold air is forced into the contained aisle, and servers draw exactly what they need.

Hot exhaust freely disperses into the larger data hall space. Hot Aisle Containment (HAC) encloses the hot aisles instead.

Google Cloud Platform extensively deploys HAC across their facilities because it's often more effective-containing the smaller volume of hot air rather than the larger cold air space.

Facebook's data centers in Prineville, Oregon use HAC with return air temperatures reaching 95-105°F, far exceeding traditional comfort cooling ranges but perfectly safe for IT equipment.

The key difference: CAC creates positive pressure in cold aisles (pushing cold air toward servers), while HAC creates negative pressure in hot aisles (pulling hot air away from servers).

Measurable Efficiency Gains QTS implemented hot aisle containment across their Atlanta-Metro facility and documented remarkable improvements:

• Supply air temperature increased from 55°F to 65°F (less cooling energy required)
• Return air temperature reached 95°F (better heat extraction)
• Overall cooling efficiency improved by 20-30%
• Eliminated nearly all thermal hotspots above 80°F These numbers matter because every degree you raise the supply air temperature saves approximately 2-4% on cooling energy costs.

For a 5 MW facility spending $500,000 annually on cooling, that's $10,000-$20,000 saved per degree.

Calculating Cooling Capacity: Sizing Your Thermal Infrastructure Cooling capacity calculations follow a straightforward principle: your cooling infrastructure must remove heat at the same rate IT equipment generates it, plus additional capacity for redundancy and future growth.

Understanding Cooling Measurement Units Cooling capacity uses two primary units: Tons of refrigeration: One ton equals 12,000 BTU/hr or 3.517 kW.

This legacy measurement originated from the cooling capacity of one ton of ice melting over 24 hours.

CoreSite's LA1 facility lists 2,400 tons of total cooling capacity. Kilowatts (kW): More commonly used in modern data centers because it directly correlates with electrical load.

If your IT equipment draws 1,000 kW, you need at least 1,000 kW of cooling capacity to remove that heat.

Conversion formula: Tons × 3.517 = kW or kW ÷ 3.517 = Tons

Basic Capacity Calculation Start with your IT load:

• Total IT equipment power draw: 2,000 kW
• Redundancy factor (N+1 configuration): multiply by 1.25
• Growth buffer (20%): multiply by 1.2 Required cooling capacity = 2,000 kW × 1.25 × 1.2 = 3,000 kW or approximately 853 tons If you're using CRAC units rated at 20 tons each, you'd need 853 ÷ 20 = 43 units (round up to 44).

With CRAH units at 75 tons each, you'd need just 12 units.

Rack Density Considerations Switch's SUPERNAP facility in Las Vegas supports rack densities up to 30 kW per rack-these high-density deployments require special attention.

A standard floor tile delivers approximately 200-300 CFM (cubic feet per minute) of airflow.

Using the simplified formula: CFM required = (kW × 3,413) ÷ (1.08 × ΔT) Where ΔT is the temperature difference between supply and return air.

For a 20 kW rack with a 20°F temperature delta: CFM = (20 × 3,413) ÷ (1.08 × 20) = 3,160 CFM That's the airflow from 10-15 standard floor tiles, which explains why high-density racks require supplemental cooling solutions like in-row cooling units or rear-door heat exchangers.

Practical Examples

Example 1: Right-Sizing Cooling for a Colocation Build-Out You're leasing 10 racks in an Equinix IBX facility.

Each rack will house networking equipment and servers totaling 8 kW per rack.

The facility provides cooling, but you need to verify they've allocated sufficient capacity.

Total IT load: 10 racks × 8 kW = 80 kW Cooling requirement: 80 kW × 1.25 (N+1) = 100 kW of cooling capacity The facility spec sheet shows 150 tons of total cooling capacity serving 40 racks.

Converting tons to kW: 150 × 3.517 = 528 kW.

Per-rack allocation: 528 kW ÷ 40 = 13.2 kW per rack.

Your 8 kW per rack fits comfortably within the 13.2 kW allocation.

You have 5.2 kW of headroom per rack for future growth-equivalent to 65% overhead.

The facility is properly provisioned for your needs.

Example 2: Troubleshooting a Thermal Hotspot Microsoft Azure's operations team notices server inlet temperatures climbing to 82°F in one section of their data hall-above the 81°F ASHRAE recommended limit.

The affected racks draw 15 kW each, and the facility uses cold aisle containment with underfloor air delivery.

Investigation reveals:

• Floor tiles in front of affected racks deliver only 150 CFM each (standard perforated tiles)
• Hot aisle temperatures measure 110°F (normal)
• Adjacent racks with 10 kW loads show normal temperatures The problem: insufficient airflow for the thermal load.

High-density racks need more CFM than standard tiles provide.

The solution deployed:

• Replace standard perforated tiles with high-flow tiles delivering 400 CFM each
• Add two additional floor tiles per rack (increasing total CFM)
• Install blanking panels to eliminate air recirculation gaps Result: Inlet temperatures dropped to 72°F, providing comfortable operating margins.

Example 3: Evaluating CRAC vs CRAH for a New Facility You're planning a 3 MW IT load facility.

Two architectural options: Option A: CRAC units

• 50 units at 25 tons each = 1,250 tons total
• Capital cost: $15,000 per unit = $750,000
• Operating cost: Higher PUE of 1.8 means 5.4 MW total facility power
• Annual energy cost (at $0.08/kWh): $3.8 million Option B: CRAH with central chiller plant
• 15 units at 75 tons each = 1,125 tons total
• Capital cost: $500,000 for chillers + $300,000 for CRAHs = $800,000
• Operating cost: Better PUE of 1.4 means 4.2 MW total facility power
• Annual energy cost: $2.9 million Option B costs $50,000 more upfront but saves $900,000 annually in energy costs.

The payback period is just 20 days.

Over 10 years, Option B saves $9 million in operating expenses-a compelling business case for CRAH deployment.

Common Misconceptions Misconception: "More airflow always means better cooling" Excessive airflow creates its own problems.

Imagine flooding your network with unnecessary traffic-it consumes bandwidth without adding value.

Too much airflow increases fan energy consumption (fans typically consume 10-15% of cooling system power) and can create pressure imbalances that actually worsen hot spots.

Digital Realty's engineering teams carefully balance airflow using computational fluid dynamics (CFD) modeling to deliver exactly what's needed-no more, no less.

Optimal cooling comes from proper airflow distribution, not maximum volume. Misconception: "CRAC and CRAH units work the same way, they're just named differently" This is like saying containers and virtual machines are the same because they both run applications.

The architectural differences matter enormously.

CRACs use direct expansion refrigeration cycles with compressors-they're monolithic cooling solutions.

CRAHs separate cooling generation from distribution using chilled water, enabling centralized optimization.

When Switch designs their hyperscale facilities, choosing CRAH architecture allows them to implement free cooling economizers that wouldn't work with CRAC units.

The distinction affects everything from energy efficiency to maintenance procedures to scalability limits.

Summary & Key Takeaways

• CRAC units are self-contained air conditioners with built-in refrigeration (10-30 tons each), suitable for smaller facilities, while CRAH units use central chilled water plants and scale better for large deployments (50-100+ tons each)
• Hot aisle/cold aisle containment organizes airflow by alternating rack orientations-containing either the cold or hot aisles prevents air mixing and can improve cooling efficiency by 20-30% while allowing higher supply air temperatures
• Cooling capacity calculations start with IT load in kW, add redundancy factors (N+1 = 1.25×), and growth buffers (typically 1.2×); one ton equals 3.517 kW of cooling capacity
• Airflow requirements scale with rack density-a 20 kW rack needs roughly 3,000+ CFM of airflow, requiring specialized high-flow floor tiles or supplemental in-row cooling beyond standard raised-floor distribution
• System selection between CRAC and CRAH architectures should consider total cost of ownership, not just capital expenses-better PUE from CRAH systems typically pays back the higher initial investment within months
• Proper containment eliminates thermal mixing and hotspots, enabling facilities to operate with supply temperatures of 65°F or higher rather than over-cooling to 55°F, reducing energy consumption by 2-4% per degree

Next Steps Build on these air cooling fundamentals by exploring liquid cooling technologies, which are becoming essential for high-density AI and HPC workloads exceeding 30 kW per rack.

Understanding chiller plant design and free cooling economizers will deepen your knowledge of how CRAH systems achieve superior efficiency.

Energy efficiency metrics and PUE (Power Usage Effectiveness) calculations will help you quantify the financial impact of cooling design decisions covered in this lesson.