CFD for Data Centre Cooling: Preventing Recirculation & Hot Spots

Jan 22
5 min read

CFD Analysis for Data Centre Cooling System, London, UK, Copyright: Mansim

UK data centres are being pushed hard by higher-density racks, AI workloads, and the constant pressure to improve efficiency without compromising uptime. That combination makes thermal problems feel “sudden”: a room that used to run comfortably can develop persistent hot spots, unstable inlet temperatures, or cooling plant that underperforms on warm or windy days.

The frustrating part is that many of these issues aren’t caused by a lack of cooling capacity. They’re caused by airflow and heat going to the wrong places, especially recirculation (hot exhaust finding its way back to an intake) and mixing (hot and cold streams blending before they do their job).

This post explains what recirculation and hot spots really look like in practice, why they’re common in UK facilities, and how computational fluid dynamics (CFD) helps you identify and fix them before they become an operational headache.

What “recirculation” and “hot spots” actually mean

Recirculation

In cooling terms, recirculation is simple: warm air (or warm exhaust) returns to an intake instead of being rejected to ambient. It can happen:

Inside the data hall: hot exhaust air from racks leaks/mixes back into cold aisles, raising server inlet temperatures.
Outside on rooftops/plant areas: condenser exhaust or hot discharge plumes are pulled back into air intakes or across heat exchangers, reducing cooling capacity and increasing energy use (and sometimes triggering alarms at the worst moments). This “hot air entrainment” risk is a major reason external CFD is increasingly used in data centre rooftop mechanical design.

Hot spots

Hot spots are localised regions where IT inlet temperature exceeds your target or allowable envelope, often only affecting a subset of racks. In many cases, the room-average temperature looks fine — but a few racks are starved of cold air, or they’re ingesting mixed/recirculated air.

Industry guidance (including ASHRAE TC 9.9 thermal guidelines reference materials) exists specifically because keeping IT equipment within appropriate thermal limits is fundamental to reliability and availability.

Why this is such a UK data centre issue right now

A few practical realities make recirculation risk more acute in the UK:

Densification and retrofits: Many UK sites are evolving in phases: adding new high-density halls, retrofitting cooling plant, or swapping air-cooled equipment for hybrid/water-cooled systems. Changes in one area can create new flow paths and new failure modes (especially on rooftops with dense mechanical layouts). One example is a London data centre rooftop upgrade where equipment replacement changed the flow interactions enough to justify detailed external CFD across multiple wind conditions.
Wind-driven effects on rooftops: UK weather patterns and urban surroundings can create complex, shifting wind fields around plant screens, parapets, neighbouring buildings, and plume buoyancy. External CFD is specifically useful because it can quantify how wind and heat interact with real geometry — rather than relying on generic “rule-of-thumb” separation distances.
The efficiency–resilience squeeze: Teams want lower fan power, higher supply temperatures where possible, and better overall energy performance — but without narrowing the operational safety margin. That’s where “small” recirculation effects become expensive: a few degrees at a rack inlet can cascade into higher fan speeds, higher cooling loads, and reduced headroom.

The usual culprits (and why they’re hard to see without CFD)

Inside the data hall

Common causes of recirculation and hot spots include:

Gaps in containment (missing blanking panels, poor aisle containment, open cable cut-outs)
Poorly balanced airflow delivery (tiles/vents delivering air to low-demand areas while high-demand racks starve)
Bypass airflow (cold air short-circuiting back to returns without passing through IT)
Local pressure imbalances (one CRAC/CRAH “stealing” flow, or return paths favouring certain regions)

Containment is widely used because it reduces hot–cold mixing and makes airflow more predictable; it’s frequently discussed as a core strategy for improving thermal performance and efficiency.

Outside: rooftops and external plant

External recirculation often shows up as:

Exhaust plumes from dry coolers/chillers re-entering intakes
Warm air “pooling” behind screens or in sheltered zones
Short-circuiting between adjacent plant rows
Interactions between heat rejection equipment and other exhaust sources (for example generator/DRUPS exhaust streams)

These effects are highly geometry- and wind-direction-dependent — exactly the kind of problem CFD is built to solve.

What CFD does differently (and why it’s the right tool here)

CFD gives you a physics-based view of:

Where the air actually goes (not where drawings assume it goes)
How heat moves through the hall/plant area under realistic boundary conditions
Which design changes matter most (and which ones are noise)

This can be used in two complementary ways:

1) Internal CFD: airflow and temperature in the data hall

Use internal CFD when you need to understand:

Rack inlet temperature distribution (and worst-case inlets)
Recirculation paths from hot aisle to cold aisle
Raised-floor / overhead supply effectiveness
The thermal impact of equipment failure scenarios (e.g., fan wall/CRAH off)

2) External CFD: rooftop and site-level heat rejection performance

Use external CFD when you need to assess:

Hot air entrainment / recirculation at chillers, dry coolers, hybrid coolers
Wind effects on plume dispersion and intake temperature
The impact of barriers, screens, parapets, and surrounding structures
Layout optimisation (spacing, orientation, discharge direction, elevation)

Recent industry discussion highlights that external CFD fills a critical gap: it explicitly captures wind–plant–geometry interactions that internal hall models do not.

A practical CFD workflow for preventing recirculation and hot spots

Here’s a simple, repeatable approach that works well for both design and troubleshooting:

Step 1: Define the “decision” (not just the simulation)

Before modelling, agree what you’re trying to decide:

“Which racks are at risk at peak load?”
“Will this new plant layout recirculate under dominant wind directions?”
“Does adding a screen/barrier reduce intake temperature enough to matter?”
“How much headroom do we have before we violate our inlet temperature targets?”

ASHRAE guidance is often used as a reference point for environmental envelopes and operational intent, so it’s useful to align internal stakeholders on what “acceptable” means in your context.

Step 2: Build the right level of geometry (don’t overcomplicate)

For internal CFD: focus on racks, aisles, containment, supply/return paths, and major obstructions.
For external CFD: include all heat rejection equipment, relevant screens/barriers, key surrounding buildings, and exhaust sources that can interact.

Step 3: Simulate the scenarios that drive risk

For the UK, that typically includes:

Multiple wind directions and representative wind speeds (external)
Seasonal extremes / peak ambient conditions (external)
Peak IT load distribution and partial load cases (internal)
Failure and degraded-operation scenarios (internal and sometimes external)

Step 4: Report in “operator language”

Good CFD outputs aren’t just pretty flow plots. They should answer:

Where are the worst-case rack inlets?
Which equipment/intakes see elevated temperature (and by how much)?
Which flow paths cause the problem?
Which interventions reduce risk the most, and what’s the cost/complexity trade-off?

Mitigation strategies that typically work

Inside the hall

Containment improvements (seal leaks, improve doors/roof panels, fix bypass routes)
Airflow balancing (tile/vent distribution, underfloor pressure management)
Rack hygiene (blanking panels, cable management, sealing unused openings)
Targeted supplementary airflow (only where the CFD shows starvation)

On rooftops / external plant

Layout changes (spacing, staggering, orientation to dominant winds)
Barriers and wind screens (used carefully; they can help or worsen pooling depending on geometry)
Discharge direction changes (avoid discharging into sheltered zones)
Elevation or separation (reduce short-circuiting between rows)
Operational controls (staging, sequencing, and setpoints informed by “risk conditions” revealed by CFD)

External CFD is particularly valuable here because mitigation measures can have unintuitive effects—especially once wind and buoyancy interact with screens and parapets.

A quick “pre-CFD” checklist (useful even before you model)

If you’re seeing hot spots or unstable temperatures, capture these before you start:

Rack-level inlet temperatures (at least for the worst zones)
IT load maps (kW per rack, and how it shifts)
Cooling unit flow rates and setpoints
Containment condition survey (photos + notes)
Known failure events (what happened, when, and what alarms triggered)
Rooftop photos/layout drawings (including screens, parapets, neighbouring structures)

The goal is to make CFD targeted: model the scenarios that actually match your risk profile.