Liquid Cooling Responsibilities for Colocation Providers

Draw where the fluid goes, then assign the work#

A rack alarm does not identify the owner of the coolant or the operator of an upstream valve. Start with the installed drawing. In a common liquid-to-liquid arrangement, the CDU heat exchanger separates the facility water system (FWS) from the technology cooling system (TCS) that feeds the server cold plates.

Record the fluid paths, equipment, isolation points and dependent loads. Then add ownership and contracted service. The provider might maintain a tenant's CDU or own a CDU serving several customers; either arrangement needs explicit operating responsibilities.

A shared cabinet does not necessarily mean shared coolant. Identify separate heat exchangers, pumps and circuits before deciding whether a fluid change can affect another tenant.

Finally, mark the contractual delivery point and its measurement method. Service might be defined at a flange, rack manifold or specified cooling condition. The Open Compute Project (OCP) distinguishes the TCS fluid program from the facility side; Vertiv describes several ways to divide facilities and information technology (IT) responsibilities.

References: Open Compute Project: Water-Based Transfer Fluid Guidance: TCS and FWS Scope; Vertiv: Liquid Cooling Services: IT and Facilities Responsibilities

Make the cooling promise measurable#

Terms such as 'liquid cooling ready' leave important decisions open. Agree the actual supply conditions, capacity, measurement points and normal control behavior. Describe what happens when the service moves outside its permitted operating range.

Include planned maintenance: what capacity or redundancy will be unavailable, which tenants depend on it and who approves the window. Assign approval for new server generations whose requirements differ from those originally accepted.

Acceptance needs evidence for both infrastructure and tenant connections. Keep fill and cleanliness records, equipment readiness, approved fluid, alarm tests, control state and the operating baseline. Make clear which tests cover the tenant's circuit and which cover only the upstream plant.

References: ASHRAE: Commissioning and Performance Validation

Put a decision owner against each activity#

A RACI matrix makes the division of work readable: Responsible, Accountable, Consulted and Informed. Use one accountable decision owner for each row, alongside the team that performs the work. The same party can hold both roles.

In the table, A approves or owns the decision, R executes, C is consulted beforehand and I receives the update. Writing 'shared' without naming a decision owner leaves the difficult handoff unresolved.

For provider-owned FWS and tenant-owned CDU and TCS, the matrix below assigns fluid work to a specialist and equipment advice to the original equipment manufacturer (OEM). Confirm the assignments against the service agreements. A provider-owned shared CDU needs revised rows and participation by dependent tenants in acceptance and outage approval.

Connect each alarm to an authorized responder#

A tenant engineer may see a CDU alarm while only provider personnel can access an upstream isolation valve. That arrangement can work when the escalation route is clear and tested. It fails as an operating model when neither party knows who has authority to act.

Name who can acknowledge alarms, view trends, enter the rack area, change setpoints, operate valves and authorize fluid work. Receiving an alert should lead to a defined contact and action rather than a search for the equipment owner.

Share circuit and asset IDs, event time, state, trend quality, associated load and service history. Agree a time basis and the system retaining original records. Test responders' read access or export process during an incident exercise.

For shared service, register affected tenants, escalation contacts and permitted information sharing. The coordinator needs to identify dependent loads while respecting tenant access boundaries. Vertiv identifies shared visibility and escalation as important parts of the operating model.

• Name primary and backup responders for mechanical, IT and fluid issues.
• Record access restrictions and the method for emergency entry or remote support.
• Test alarm delivery, acknowledgement, escalation and data access together.

References: Vertiv: Liquid Cooling Services: IT and Facilities Responsibilities

Agree the response while the system is healthy#

A chemistry advisory, loss of flow and a suspected leak call for different decisions. Define each alarm class, the evidence needed for escalation and its project-approved response before a live event.

Identify who may isolate equipment, protect a load or coordinate a controlled compute stop. The approved procedure must account for the actual circuit and protection sequence; a blanket shutdown instruction cannot resolve those differences.

Document protection steps, communication order, authority and evidence preservation. Coordinate authorized IT workload actions with mechanical valve and plant actions so an intervention does not unexpectedly remove cooling from an operating load.

Practice a scenario with unavailable contacts and incomplete telemetry. Check that a backup can make the required decision and that responders understand which readings are stale. The exercise should end with a return-to-service decision, not only acknowledgement of the first alarm.

Incident and circuit IDs: [IDs]. Alarm class: [event]. Current evidence: [values, quality and time]. Affected loads: [dependency register]. Incident coordinator: [name]. Authorized mechanical action: [procedure and owner]. Authorized IT action: [procedure and owner]. Approvals and notifications: [record]. Recovery evidence: [tests]. Return-to-service decision: [accountable owner].

One sample, three tenants: scope the service window#

For a provider-owned CDU serving three tenant branches on one TCS circuit, an out-of-range coolant sample needs a coordinated review. Even if only one tenant reports a symptom, a common fluid intervention could affect all three branches.

Use the agreed incident coordinator to contact all three tenants and arrange confirmatory tests with the fluid specialist. The circuit drawing identifies who could be affected; the operating agreement identifies who may authorize the work.

Verify history and sampling before choosing an intervention. If confirmatory tests support a fluid correction, obtain approval and agree the permitted compute state with each tenant. Record the procedure, quantities, batches and resulting tests. The designated owner accepts return to service when recovery criteria are met.

Keep the agreement useful after handover#

Warranty and service coverage follow the actual contracts and supplier terms. Record the required fluid, materials, service qualifications, tests and notifications for the deployed equipment. A guideline or monitoring installation does not itself establish coverage; proposed changes need review by the party authorized under those agreements.

Keep a handover pack that a new responder can use: the circuit register, responsibility matrix, approved procedures, fluid records, commissioning evidence, spares, alarm routes and contacts. Name who retains samples and laboratory reports, and how the relevant parties access them after a staff change or equipment refresh.

Review responsibilities after hardware, capacity, formulation, topology or contractor changes. Date and approve the revised document, update contacts and retain the associated training record.