OSDCTC.022 asks a reliability question: when a coolant distribution unit (CDU) loses its operating pump, can the remaining equipment still protect an AI rack? A spare pump improves resilience only when its available flow, pressure, power supply, and controller response match the actual cooling requirement. After OSDCTC.021 on balancing coolant branches, the next skill is interpreting redundancy as a tested capability rather than counting pump motors.
Objectives and prerequisites
- Define N, N+1, duty/standby, and failure domain.
- Distinguish pump-start indication from proven flow through critical IT branches.
- Identify the telemetry and acceptance records required for a failover assessment.
- Recognize the safety limits of commissioning and failure testing.
Prerequisites: read OSDCTC.016 on CDU monitoring and understand basic hydraulic flow, differential pressure, and rack cooling requirements.
What N+1 actually promises
N is the minimum capacity required to carry the specified design load. N+1 adds one independently usable capacity element beyond that requirement. For an illustrative system needing two equivalent pumps to maintain an approved flow target, three properly sized pumps might provide N+1 resilience: after losing one, two can still serve the load. But if all three are necessary at peak demand, the arrangement has no spare pumping capacity at that condition. Equipment count alone proves nothing.
The Open Compute Project’s liquid-to-liquid CDU testing guide identifies pump redundancy as important for continuity during repair or malfunction. Google’s Project Deschutes description likewise identifies redundant pump and heat-exchanger components in deployed CDU architecture. Neither reference guarantees a specific site’s installed system passes a failover test.
Why a standby start is not the finish line
A pump can show RUN while system cooling remains inadequate. A closed valve, clogged strainer, air pocket, failed check valve, reduced pump head, or incorrect control mode may leave cold plates undersupplied. Operators must compare measured circulation and pressure against the approved system boundary. A temperature sensor may lag the underlying hydraulic event; an apparently normal coolant temperature at the first moment does not establish successful recovery.
Fault domains also matter. Two pumps sharing one vulnerable controller, electrical feed, or common suction path are not automatically independent. The data center reliability engineering lesson explains why maintenance and failure analysis consider shared dependencies.
Read-only evidence checklist
- Record the CDU identifier, design load, approved pump arrangement, controller revision, and active alarm history.
- Review synchronized pump status, speed, electrical faults, coolant flow, differential pressure, supply and return temperatures, and leak alarms.
- Compare historical flow and device temperatures under equivalent IT loads, noting changes during previously authorized maintenance events.
- Verify whether standby equipment has an independent power path and whether the building management system records the correct alarm sequence.
- Locate the signed commissioning test report and acceptance criteria. Escalate missing proof rather than claiming untested redundancy.
Cross-check OSDCTC.020 on measuring heat removal and OSDCTC.019 on filter condition. A successful failover must preserve required cooling, not merely restore a dashboard’s green icon.
Safety and failure-test boundary
Do not disable a live pump, shut an isolation valve, change a VFD setpoint, disconnect a controller, or bypass leak protection to demonstrate failover. Only qualified teams operating under approved change control and a documented method of procedure should conduct controlled fault injection. The IT load’s allowable cooling interruption and protective response are equipment-specific; no universal safe interruption duration can be assumed.
Exercise and answer key
Scenario: A rack requires two pumps at full design load. Three equal-capacity pumps are installed, yet a shared supply breaker serves all three. Is the arrangement fully failure-independent? Answer: No. The spare pump may cover a single mechanical pump failure, but the common breaker remains a single failure domain. Verify capacity following a pump loss, electrical segregation, and the approved alarm-and-recovery sequence.
Knowledge check: Does a successful standby-pump start prove N+1 performance? Answer: No. Confirm the required branch flow and thermal response under the defined load. OSDCTC.017 on condensation controls covers another boundary that must remain protected during cooling transitions.
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiative, please scan the QR code below.


Leave a Reply