Server Room: Planning Power and Cooling for SMBs the Right Way

A server room is not a data center — and yet the same laws of physics apply. As soon as a mid-sized company consolidates more than two or three servers, a storage system and active network gear into one room, what used to be a storage closet becomes critical infrastructure. The typical scale we advise at DATAZONE is between 5 and 15 kW of IT load — roughly 1 to 3 well-populated 19-inch racks. This is exactly the range where most planning mistakes happen: cooling units too small, UPS too weak, airflow left to chance. This article shows how to realistically size power and cooling without overshooting your target.

Load Calculation: What Actually Draws Power?

Before you order a cooling unit or a UPS, you need an honest number: the actual electrical load of your IT. Datasheets only help so much, because the values printed on them almost always list the power supply’s maximum rating — not the operating point.

The pragmatic approach is an inventory from three sources:

1. Measurement at the PDU or UPS output over at least a typical business week. Include peaks from backup windows, batch jobs or ERP month-end runs.
2. IPMI/Redfish polling of the servers (Dell iDRAC, HPE iLO, Supermicro IPMI). Current wattage can be pulled cleanly via SNMP or Redfish.
3. PSU nameplate rating × 0.4 to 0.6 as a rough rule of thumb for servers in normal operation when no measurement is possible.

A typical example calculation for a small Proxmox cluster with TrueNAS looks like this:

Component	PSU Nameplate	Realistic Continuous Load
3× Proxmox node (2× 800 W redundant)	4,800 W	approx. 1,200 W
1× TrueNAS R30 with 24 HDDs	2× 1,200 W	approx. 700 W
1× Proxmox Backup Server	2× 550 W	approx. 250 W
Core switch + ToR switch	2× 250 W	approx. 220 W
Firewall HA pair (OPNsense)	2× 250 W	approx. 90 W
KVM, monitoring appliance, misc.	—	approx. 100 W
Total IT load	—	approx. 2.6 kW

On paper this would total more than 11 kW. In reality it is less than 3 kW. Exactly this difference decides whether you buy a 5 kW UPS or unnecessarily oversize to a 15 kW unit. Still, plan for a safety margin of 30 to 50 percent on top of the measured load — for expansion, peaks and aging.

Redundancy Concept: N+1, 2N or Nothing?

The second fundamental decision is the redundancy tier. In practice we encounter three models:

• N (no redundancy): One cooling unit, one UPS, one power feed. If it fails, the server room goes down. Defensible for pure test or development environments, not for productive SMB infrastructure.
• N+1: One redundant unit in addition to the calculated demand. Two cooling units where one alone can carry the load. Two UPS modules where one may fail. The standard model for the mid-market.
• 2N: Everything doubled, including separate feeds from the building supply and separate cooling loops. For high-availability environments with RTO close to zero. Significantly more expensive, usually only justified above 20 kW.

An honest recommendation: For most SMB server rooms in the 5 to 15 kW range, N+1 is the economic sweet spot. 2N doubles the investment but addresses failure scenarios that can never be fully decoupled inside a single room anyway (fire, water, smoke). If you truly need 2N availability, think about a second site — not about duplicated cooling units in the same room.

Sizing Cooling: From kW to BTU/h

Cooling follows a simple rule: every watt of electrical power that enters the server room comes out again as heat. If you run 2.6 kW of IT load, you also need to remove 2.6 kW of heat — plus headroom for people, lighting, solar gain and reserve.

Conversion between the common units:

1 kW  = 3,412 BTU/h
2.6 kW ≈ 8,870 BTU/h
5 kW  ≈ 17,060 BTU/h
10 kW ≈ 34,120 BTU/h

In practice we apply a factor of 1.3 to 1.5 to the IT load: For 2.6 kW of IT we size at least 3.5 to 4 kW of cooling capacity — per unit under N+1. Two units at 4 kW each, where one alone carries the load.

For the technical implementation, three options are available in the SMB space:

• Split air conditioners (comfort cooling): Cheap, but not designed for continuous duty. We see them regularly in small server rooms and just as regularly they fail in mid-summer. Only acceptable as a temporary solution.
• Precision cooling units: Built for continuous operation, with defined airflow, filter stages and redundancy sensors. Typical starting point for serious server rooms from around 5 kW.
• In-row cooling: Cooling unit directly between the racks, very short air paths, high efficiency. Makes sense at higher densities (from around 8 to 10 kW per rack) or in cramped rooms where classical room-level airflow does not work.

For 5 to 15 kW total load, a pair of precision cooling units is usually the right choice. In-row becomes interesting as soon as you load individual racks with more than 6 to 8 kW, as happens with dense virtualization hosts or GPU nodes.

Airflow: Why Hot/Cold Aisle Works Even at Small Scale

A common mistake in small server rooms: the cooling unit blows into the room, servers pull in air, no defined separation of cold and hot air. The result is short-circuit circulation — hot exhaust is immediately pulled back in, servers run hot, the AC runs flat out.

Even with only one or two racks, basic order pays off:

• Install all servers facing the same direction. Cold air in the front, hot air out the back. Retrofit exceptions like rear-breathing switches with reverse-airflow kits.
• Blanking panels in every empty rack unit. Without panels, hot air flows through the rack from back to front and contaminates the cold air.
• Floor and rack seals at cable openings. Every open U is a short-circuit path.
• Cold aisle or hot aisle containment from two to three racks onwards. Ready-made kits are available in the mid four-digit range and typically amortize through cooling savings within a few years.

ASHRAE recommends a cold-aisle temperature range of 18 to 27 degrees Celsius at 20 to 80 percent relative humidity for IT rooms in class A1. The widespread belief that a server room has to be cooled down to 20 degrees is outdated — every additional degree of return air temperature saves cooling energy.

UPS Integration and Monitoring

The UPS takes on two jobs: bridging short power outages and enabling a clean, coordinated shutdown when power stays off longer. For a typical 5 kW server room we plan around the following:

• Autonomy time: 10 to 20 minutes at full load are enough for a controlled shutdown of VMs. Anyone wanting to bridge longer needs a generator — the UPS alone cannot do this economically.
• Topology: Online (double-conversion) UPS for server room operation. Line-interactive units belong at the workstation, not on storage.
• Modularity: Modular rack UPS can be extended later (more power modules, more battery modules) — useful if the load will grow.
• Battery replacement: VRLA batteries after 4 to 5 years, Li-Ion after 8 to 10 years. A battery test belongs in the maintenance contract.
• Shutdown integration: NUT (Network UPS Tools) or vendor agents (APC PowerChute, Eaton IPP) on every Proxmox host. Automatic shutdown of VMs before UPS depletion — otherwise the best UPS is useless.

Monitoring is the second half of the story. Without sensors you often notice a cooling failure only when the first servers shut themselves down thermally. A baseline setup includes:

• Temperature and humidity sensors in the cold aisle and hot aisle of every rack (at least top and bottom)
• Water sensors under the cooling unit and in the raised floor
• Smoke detectors with alarm forwarding
• Mains power monitoring via PDU with SNMP
• UPS status via SNMP or Modbus into the central monitoring

We usually integrate such sensors via Linux-based monitoring systems such as Zabbix, Checkmk or Prometheus with Alertmanager. Alerts go out via email, SMS and ideally via on-call rotation to the responsible team.

Conclusion

A server room in the 5 to 15 kW range is manageable — provided that planning, redundancy and monitoring work together. The most common mistakes do not lie in exotic edge cases but in the basic design: unrealistic load assumptions, undersized cooling, missing airflow discipline and blind spots in monitoring. Anyone who measures the real load, sizes for N+1 instead of 2N and combines precision cooling with a clean hot/cold aisle separation gets away with significantly smaller investments than the catalog suggests — and still has an environment that survives summer, the backup window and the next expansion.

---

DATAZONE supports you in designing server rooms for virtualization, storage and backup — from load measurement through cooling and UPS sizing to monitoring integration. Get in touch if your infrastructure is growing and your server room needs to grow with it.