Sizing and layout studies of a data center: several crisis scenarios tested to validate the resilience of the air-conditioning systems.
EOLIOS Engineering carried out the sizing and layout studies of a data center. Several crisis scenarios were studied in order to test the resilience of the air-conditioning systems.
In brief. Sizing and layout studies of the DC10 data hall (client Equinix, Saint-Denis) through CFD simulation. The first studies were not compatible with the ASHRAE standards governing the site and a hot-spot risk was identified: some floor grilles close to the walls were overflowing, under-sizing decisive grilles near a high-power server. After optimising the grille positioning and verifying the absence of hot spots, several scenarios were studied, up to the evolution of temperatures in the event of a utility-grid outage and reactivation of the backup systems.
In the very-large-scale data center environment, cooling-system standards able to meet the evolving needs of IT are required, because of the growing density of the equipment (more than 10 kW / rack). Removing the heat from the high-density equipment is a key point.
This means numerically simulating the thermal-airflow behaviour of the various phenomena occurring in the hall, in connection with the IT processes and the air-conditioning systems.
At the heart of the mission is the construction of a digital twin of the data hall: a faithful virtual replica of the room, its cabinets and all the air-handling systems. This model is built from the site drawings, the 3D mock-ups and the equipment datasheets, from which the supplied air flow rates, the power dissipated per cabinet and the temperature set-points are extracted.
Once fed with this data, the digital twin reproduces the thermal-airflow behaviour of the room and serves as a virtual test bench: layout configurations, operating scenarios and crisis situations are tested on it without ever disrupting production.
A virtual, dynamic representation of a real installation, fed by its design and operating data. In a data center, it makes it possible to replay the behaviour of air and temperatures by computation, to anticipate hot spots and to test failure scenarios before they occur.
A server-room layout principle consisting of alternating cold-air intake aisles and hot-air exhaust aisles. It avoids the mixing of flows, limits recirculation and directly conditions cooling efficiency.
Computational fluid dynamics (CFD) divides the room's air volume into a mesh of computation cells, then solves the flow and heat-transfer equations at each point. It thus reproduces the air velocity, the temperature and the pressure at every point of the data hall, from the floor grilles to the cabinet intakes.
Survey and modelling: room geometry, cabinet layout, supply and return systems.
Meshing: division of the air volume, refined at the grilles and cabinet intakes.
Boundary conditions: supplied flow rates, power dissipated per cabinet, air-conditioning system set-points.
Exploitation: temperature and velocity maps, verification of compliance with the ASHRAE ranges and identification of hot spots.
ASHRAE recommendations framing the admissible temperature and humidity ranges at the intake of IT equipment. They serve as a reference for judging whether a cooling configuration is acceptable.
The results of the first studies were not compatible with the ASHRAE standards governing the site, and a hot-spot risk was identified. Some floor grilles close to the walls could show an overflow, leading to an under-sizing of decisive grilles located near a high-power server.
After optimising the positioning of the grilles and checking the absence of hot spots, several scenarios were studied.
Key takeaway. The correct positioning of the supply grilles is decisive: a local overflow can under-supply a critical cabinet and create a hot spot.
In a raised-floor room, the space beneath the tiles acts as a supply plenum: cold air is pressurised there by the precision air-conditioners, then delivered into the cold aisles through perforated grilles or adjustable-flow tiles. Cooling performance depends less on the installed cooling power than on the distribution of that air where the servers need it.
Two quantities govern this distribution: the static pressure in the plenum and the perforation ratio (or opening) of the tiles. A grille placed too close to an air-conditioner sits in a zone where the air moves fast and the static pressure is low, which can reduce its flow, or even cause a reverse suction that draws warm air back down into the plenum. Conversely, a grille placed at the end of the plenum, where the pressure has recovered, may end up in overflow at the expense of its neighbours.
The volume beneath the raised-floor tiles, pressurised by the air-conditioners and used to distribute cold air towards the aisles. Its height, its clutter (cables, pipes) and the grille layout determine the pressure available at each point.
The proportion of open surface of a raised-floor tile (perforated or damper grille). It sets the air flow delivered for a given plenum pressure: too high near the air-conditioners, it unbalances the distribution; well distributed, it aligns the flow with the thermal load of the cabinets.
The challenge is to align the cold-air flow with the thermal load of each cabinet, while preserving a homogeneous plenum pressure. CFD simulation makes it possible to objectify these choices, cabinet by cabinet:
Position the grilles in the cold aisles only, facing the servers' air intakes, never in the hot aisles where they would waste cold air.
Avoid the immediate vicinity of the air-conditioners, where the low static pressure degrades, or even reverses, the flow of the first grilles.
Dose the perforation ratio and use adjustable-damper tiles to boost the flow facing high-power cabinets.
Clear the plenum of obstacles (cable trays, pipes) that create pressure losses and low-pressure zones.
Contain the overpressure at the end of the row to avoid the overflow of the distant grilles at the expense of the decisive ones.
Key takeaway. A raised-floor grille only delivers useful cooling if the plenum pressure and its perforation ratio are consistent with the load of the cabinet it serves. Correct placement, validated by CFD, prevails over adding cooling power.
Several scenarios were simulated: standard-use scenarios and failure or maintenance scenarios. Finally, we studied the evolution of the room temperatures over time in the event of a utility-grid outage and the reactivation of the various backup systems.
Where a steady-state simulation describes the room stabilised at a given instant, the transient study computes the evolution of temperatures second by second during an event. It is the only way to answer the real resilience question: in the event of a loss of cooling, how much time is available before the servers exceed their critical thresholds?
The sizing scenario is the utility-grid outage. At the moment of the switchover, the precision air-conditioners stop while the generator sets start up and take over the load, whereas the servers keep dissipating their full power. The room air temperature then rises very fast, all the more so as the thermal inertia available in the plenum and the air volume is low compared with the densities involved.
An unsteady simulation that solves the evolution of the temperature and velocity fields over time, from an initial state and a chronology of events. It quantifies the rates of temperature rise and the delays before thresholds are crossed, where a steady-state study only gives a frozen state.
The capacity of the masses present (air, structures, equipment) to dampen a temperature variation. Low in a dense data hall, it leaves little margin: a few tens of seconds are sometimes enough to reach the thresholds if cooling stops.
The study reconstructs the entire recovery timeline: cooling shutdown, generator start-up delay, fan restart and then progressive restoration of the set-points. By following the evolution cabinet by cabinet, the simulation highlights the zones that reach their limits first and verifies that none crosses the thresholds before cooling returns.
Rate of temperature rise per zone, from the moment of the outage.
Delay before critical threshold compared with the switchover time of the backup systems.
Recovery kinetics back to the set-point after restart of the generators and fans.
Most exposed zones, to be treated as a priority through layout or redundancy.
Key takeaway. The resilience of a data hall is not judged at equilibrium but in the transition: the transient study compares the delay before the critical threshold with the recovery time of the backup systems, and proves the room's endurance before the real incident.
By replaying each configuration through computation, the digital twin makes it possible to make the sizing and layout of the room reliable before any physical intervention. Corrections, such as repositioning the grilles, are validated virtually, which reduces the risk of error in operation.
Anticipation of hot spots and verification of compliance with the ASHRAE ranges.
Testing of crisis scenarios (failure, maintenance, grid outage) without risk to production.
Optimisation of the layout of the cabinets and the supply for better cooling efficiency.
A durable decision-support tool: the model can be reused at each evolution of the room.
Key takeaway. The digital twin turns the cooling of a data hall into a measurable problem: each scenario is tested virtually before being deployed, which secures the room's availability.
ASHRAE standards, grille positioning and crisis scenarios: the answers to the questions operators and designers ask before an internal CFD study.
To carry out the sizing and layout studies of the data hall and test the resilience of the air-conditioning systems through several crisis scenarios, by simulating the thermal-airflow behaviour of the hall's phenomena, as on our internal study of the DC28 data center.
The result of the first studies was not compatible with the ASHRAE standards governing the site and a hot-spot risk was identified: some floor grilles close to the walls ended up in overflow, causing an under-sizing of decisive grilles near a high-power server.
By optimising the positioning of the grilles, then verifying the absence of hot spots before studying the various scenarios.
Standard-use scenarios and failure or maintenance scenarios, as well as the evolution of the room temperatures over time in the event of a utility-grid outage and the restart of the various backup systems.
With densities that can exceed 10 kW per rack, removing the heat from high-density IT equipment is a key point for these data centers, which requires simulating the thermal-airflow behaviour of the hall.
Explore our expertise, projects and technical papers to go further than the FAQ.
3D CFD model of the data hall: sizing, grille optimisation and crisis scenarios (grid outage, backup reactivation).
Digital twinData center study – Data Hall and UPS rooms
Smoke controlSmoke-control engineering in a data center
HyperscaleExternal & internal CFD – Hyperscale Data Center
OptimisationCFD optimisation – Data Center
InternalData Center – DC28 – Internal
Technical roomsTechnical rooms – Data Center
CoolingCooling optimisation – Data Center
ExternalData Centers – DC15.1 & DC15.2 – External
ExternalData Center – PA 22 – External
GeneratorPressure-loss study – Generator – Data center
ExternalData Center – Paris
FireData Center – NOVEC gas
InternalData center – DC17 – Internal
ExternalData center – D14 – External
ExternalData center – DC25 & DC26 – External
InternalData center – DC25 – Internal
ExternalData Center – DC25 & DC26 – External
CFD expertise, delivered projects and technical dossiers: the whole Data Center field in one place.