A server room can go down without a single server fault. The cause is a power or cooling assumption that nobody has rechecked since the room was built. The equipment changed; the room did not. Each check below can be made in the room.
Single feed or dual feed
Dual power supplies only help if they draw from independent sources. The failure mode is two power supplies plugged into two PDUs that share one upstream circuit or one UPS. The redundancy exists on paper and disappears at the breaker. Confirm from the electrical schedule, or with the facilities team, that the A and B feeds are separate back to their source. Then check that dual-corded devices have both cords connected, and that single-corded kit such as a switch sits on the intended feed.
UPS runtime and battery age
A UPS estimates runtime from its current load and its battery model. Both drift. Load grows as kit is added, and batteries lose capacity with age and heat, often without an alarm until a self-test fails. Ask when the batteries were installed, when a runtime test last ran under real load, and whether that runtime still covers a clean shutdown or a generator start. If nobody knows the installation date, treat the batteries as unknown rather than healthy.
PDU capacity headroom
A rack PDU has a rated capacity per phase and often per bank. Racks filled over several years often have load stacked on whichever outlets were free, leaving one phase near its limit while another sits lightly used. The problem appears when a failed feed pushes its load onto the surviving side and trips it. Read the per-phase or per-bank figures from the PDU, and check that either side could carry the whole rack alone. Where nobody local can do this, data centre support engineers can take the readings and check labels against your records.
Hot and cold aisle discipline
Airflow works only if cold air reaches the intakes and hot exhaust does not loop back. Discipline erodes through small changes rather than design faults:
- Missing blanking panels in empty rack units.
- A device mounted back to front, or a switch with reversed airflow.
- Cable bundles blocking perforated tiles or rear doors.
- Boxes and spare kit stored in the cold aisle.
Ambient monitoring
A single thermostat on the wall tells you about the wall. Useful readings come from sensors at the rack intakes, at more than one height, with humidity recorded alongside temperature. Log them so a slow rise shows before it becomes an alarm. Confirm that alerts reach a monitored mailbox or on-call rota rather than a former employee, and test one end to end.
Small rooms at their limit
Small server rooms can run at their limit for years unnoticed. The cooling unit was sized for the original load, the UPS was bought for the original rack, and each addition since has used a little more margin. Nothing fails until a hot afternoon, a maintenance visit that switches off one cooling unit, or a mains dip. Ask: if one cooling unit, one feed or one UPS stopped now, what would the room do? If the answer is a guess, that is the first thing to measure.
