Cooling Capacity

Back to Data Center Management · Rack Unit Tracking · Power Circuit Headroom · Growth Forecasting · Service Offerings

Cooling capacity is checking that added heat load stays within what the room's cooling can actually handle. Where sensor coverage tells you what the temperatures are now, this is the question asked before anything is installed: can the heat this will produce be removed, at this position, under the conditions that matter.

1. Heat Load Is Just the Power Figure

Essentially all electrical power drawn by IT equipment leaves as heat. A rack drawing 7 kW produces 7 kW of heat, and the cooling must remove it continuously. This makes the arithmetic simple and ties this page directly to power circuit headroom — the same number answers both questions, which is why they should be asked together rather than by different people a week apart.

2. Room Capacity and Rack Delivery Are Different Numbers

This is the distinction that matters most, and the one the illustration above is built around. The room has 60 kW of spare capacity. The rack in question can be given 8 kW and is at 7.6. The new server needs 0.9 and there is 0.4 available at that position. The room's headroom is real and completely irrelevant to the question asked.

Cooling is delivered locally, and delivery is limited by how much cold air can physically reach the rack's intake:

• Airflow is usually the binding constraint, not refrigeration. The plant can chill plenty of air; the question is whether enough of it arrives at this rack. Perforated tile placement and open area, underfloor obstructions, and distance from the cooling unit all decide that.
• Recirculation steals capacity. Hot exhaust finding its way back to an intake — over the top of a rack, through a gap where a blanking panel is missing, around the end of a row — raises the intake temperature without any change in load.
• Containment changes the number substantially. Contained aisles raise the deliverable load per rack because supply and return stop mixing.
• Density has a ceiling per cooling architecture. Raised-floor rooms deliver a few kilowatts per rack comfortably; getting well beyond that generally needs in-row or rear-door cooling, which is an architecture change rather than a tuning exercise.

Which is why the answer to "we have spare cooling" is frequently "yes, and not where you want to put this". Sometimes the fix is cheap — a tile moved, blanking panels fitted, a cable bundle cleared from under the floor — and finding that out is worth the survey on its own.

3. Capacity With a Unit Down, on the Hottest Day

A single capacity figure quoted for ideal conditions is not a planning number. Two corrections are needed, and both reduce it.

The honest capacity figure is the one that holds with one unit down on the hottest day of the year. Anything higher is a figure that works until the two things coincide, which they eventually will, because a hot day is when a cooling unit is most likely to fail.

4. Signs You Are Nearer the Limit Than the Number Suggests

• Intake temperature rising with load flat — the cooling is losing ground. This is exactly the pattern trending is for.
• Delta-T falling across equipment, meaning air is moving but not through the right path.
• Cooling units running continuously at full output rather than cycling, which leaves nothing in reserve.
• A wide spread between racks. If one rack is eight degrees warmer than its neighbour, the room average is hiding a local problem and the capacity figure derived from it is optimistic.
• Supply temperature creeping up at the cooling unit despite unchanged settings.

5. Before You Buy More Cooling

Capacity problems are often distribution problems, and distribution is much cheaper to fix.

1. Blanking panels in every empty unit. The cheapest capacity available, and routinely missing.
2. Seal the floor — cable cut-outs with brush grommets, gaps at the perimeter, under cabinets.
3. Tile placement matched to actual load rather than to where racks used to be.
4. Clear underfloor obstructions, which are usually abandoned cabling.
5. Containment, even partial, before adding plant.
6. Move the load. Spreading high-density equipment across racks is free if the space and power exist, and rack unit tracking is what tells you whether they do.

How We Approach It

1. Establish total installed capacity, then the figure with one unit down at design-day ambient. That second number is the one used for planning.
2. Measure load per rack from the power data, since heat load and power draw are the same figure.
3. Determine deliverable capacity per rack, which is an airflow question and needs a survey rather than a calculation.
4. Find where delivery, not capacity, is the limit, and cost the distribution fixes before any plant is considered.
5. Answer the specific install question: can this rack take this heat, and if not, which rack can.
6. Feed the figures to growth forecasting, since cooling is frequently the constraint that binds first.

What You Get

• An honest room capacity figure: with one unit down, at design-day ambient, not the brochure number.
• Deliverable heat load per rack, which is the figure that actually answers install questions.
• A list of racks where delivery rather than capacity is the limit, with the distribution fixes costed.
• The cheap wins identified first — blanking panels, floor sealing, tile placement, underfloor clearance.
• A per-rack headroom number that can be checked in a minute when the next server is proposed.

The failure this prevents is specific and expensive: equipment installed into a rack that cannot cool it, which does not fail immediately. It throttles, it runs hot for months, and it shortens the life of everything around it.