OSDCTC.021 begins with a puzzle: a coolant distribution unit (CDU) reports adequate total flow, yet one liquid-cooled server runs hotter than its neighbors. The pump may be healthy while the rack has a distribution problem. Total liters per minute describe the combined circuit, not every cold plate. After OSDCTC.020 on measuring heat removal, the next skill is understanding where coolant travels.
Why sufficient CDU flow can hide a starving branch
Parallel server connections share supply and return manifolds. Each branch has resistance from tubing, cold plates, bends, quick-disconnect fittings, and balancing devices. A more restrictive branch generally receives less flow under the same available pressure difference. Flow meters, temperature sensors, and control valves help identify uneven distribution, but readings require known sensor locations. The Open Compute Project Cooling Environments program treats cold plates, manifolds, connectors, and CDUs as a connected cooling system.
Pressure and pressure drop are not interchangeable. The difference between branch inlet and outlet pressure drives circulation; restrictions consume available pressure. A high CDU discharge reading alone cannot prove adequate flow through the least-favored node. Pump curves, bypass paths, and valve positions matter. OSDCTC.016 explains which CDU telemetry fields operators should record.
Worked example: two unequal flow paths
For simplified turbulent flow, approximate branch pressure loss as ΔP = K × q², where K represents a branch’s resistance and q is volumetric flow. Suppose two parallel branches share equal ΔP but one has four times the resistance coefficient. Under the same fluid assumptions, the more restrictive branch carries half the flow. If total flow is 120 L/min, the model divides into 80 and 40 L/min, not 60 and 60. Real networks require vendor pressure-flow curves; laminar flow and automatic controls can behave differently.
Balanced does not always mean equal
Servers with different heat loads or cold plates may need different flow targets. Balancing means achieving each approved operating range, not enforcing identical L/min. Compare flow, supply temperature, return temperature, differential pressure, and IT load over matching intervals. The ASHRAE data-center handbook addresses liquid-loop pressure drop and heat-transfer requirements. A warm return line alone does not prove restricted flow.
Read-only troubleshooting sequence
Record rack ID, affected tray, alert, and maintenance history. Capture CDU total flow, coolant supply and return temperatures, pump state, differential pressure, and available branch telemetry, with sensor locations and timestamps. Compare readings with commissioning baselines and manufacturer flow requirements. Falling branch flow combined with rising device temperature is stronger evidence than a single warm reading.
Review trends for filter pressure drop, pump-control changes, and approved maintenance events. OSDCTC.019 covers filters; OSDCTC.018 covers connectors. Consult OSDCTC.017 when temperature changes may increase condensation risk. Never compare measurements from different fluid-circuit boundaries without qualification.
Limits and safe practices
A blockage, trapped air, malfunctioning valve, or incorrect configuration can produce similar symptoms. Never throttle an operating branch, loosen a pressurized connector, bypass a leak sensor, or change pump settings merely to test a theory. Qualified personnel must follow approved isolation, pressure-control, and commissioning procedures before physical work. A calculated flow ratio cannot replace actual measurements.
Knowledge check
Question: Two branches have resistance coefficients in a 1:4 ratio and the same pressure difference under the square-law model. What is their flow ratio? Answer: 2:1, with the lower-resistance branch carrying twice as much coolant. Takeaway: Verify the required flow at each cold plate, not only the pump’s reported total.
Editor’s note: Example values are classroom assumptions, not measured rack performance or commissioning setpoints. Videos are original-source OCP recordings. The featured image is a stylized editorial illustration, not a facility photograph.
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