Loading
Preparing page…
Checklist
Use this before procurement locks or install days begin. It surfaces the constraints that decide whether a room can host the intended load—and whether operations will inherit a design they can actually run.
Most datacenter and dedicated server-room failures are design failures discovered late: PDU circuits that cannot carry the planned density, cooling assumptions that ignore heat rejection at full load, cable pathways that force temporary runs, or network designs that never accounted for management-plane separation. Fixing those after racks are live costs more than the time spent validating them on paper.
This checklist is for the planning window—after the business need is clear, before bill-of-materials freeze and install sequencing. It is not a facilities construction guide. It is an engineering gate: confirm that physical plant, power, thermal headroom, cabling, network, security, and monitoring can support the first cut of equipment and a defined expansion path.
Work through it with facilities, network, security, and the operators who will own day-two work. Where an item cannot be answered, treat that as a blocker, not a note. Unowned assumptions become outages during the first change window or the first hire who was not in the room for the build.
Walk the groups in order. Rack layout and power set density; cooling validates that density; pathways and cabling decide how network and storage can be wired; network and security define zones and access; monitoring and documentation decide whether the environment is operable after handover. Skipping ahead usually means rework when a later constraint invalidates an earlier choice.
For each item, record a concrete answer: measured value, drawing reference, named owner, or explicit deferral with a date. “TBD” without an owner is not complete. Attach the outputs—rack elevations, circuit maps, cable schedules, IP plans—to the same package so installers and operators share one source of truth.
Revisit the checklist at design freeze and again before energization. Reality will diverge from the first draft; the goal is controlled updates, not a binder that no longer matches the floor.
Map rack positions against room dimensions, door swing, aisle widths, and manufacturer service clearances. Note weight limits for raised floor or slab and any seismic or anchoring requirements before ordering frames.
Assign rack units for compute, storage, network, and reserved growth. Call out blanking panels, cable managers, and devices that need front or rear clearance. Keep a second elevation for the next expansion so density decisions stay honest.
Sum nameplate or measured loads by rack, then map to PDU outlets and upstream circuits or phases. Record redundancy intent (A/B feeds, single-corded vs dual-corded) and the maximum additional load that can be added without a circuit change.
Decide vertical or horizontal PDUs, feed labels, and how outlet IDs map to rack units and devices. Operators should be able to isolate a device without guessing which outlet is live.
Compare expected heat rejection at planned utilization against cooling capacity and airflow path (hot/cold aisle, containment, CRAC/CRAH setpoints). Note hotspots risk for dense racks and any dependency on underfloor or overhead delivery.
Set temperature and humidity targets for intake and exhaust where relevant. Place sensors where operators will trust the reading—not only at the cooling unit return.
Write the order of bringing circuits, PDUs, and racks online, including who verifies polarity, grounding, and breaker labels before IT load is applied.
Walk tray, ladder, underfloor, and wall routes. Record fill, bend radius limits, firestop locations, and any shared pathway with power. Do not order trunks until pathways are confirmed.
Confirm local code and design rules for proximity of power and data. Document crossings and barriers so installers do not invent temporary routing on the day.
Lock category or fibre type, connector form factors, colour or identifier conventions, and maximum run lengths. State which links are permanent plant vs patchable.
Agree panel IDs, port numbering, and how they map to rack, switch, and device ports. Naming must survive moves and adds without requiring tribal knowledge.
Produce schedules that list source, destination, media, length, and label IDs for backbone and high-priority links. Leave a process for recording field changes during install.
Do not consume 100% of tray or panel ports in the first cut. Document reserved capacity and the trigger for adding plant before the next expansion.
Decide label format, print method, and who updates the as-built when a run changes. Temporary handwritten labels without a record are defects.
Document core/access or leaf roles, uplink speeds, redundancy, and spanning or fabric behaviour. State what fails when a single uplink or switch is lost.
Segment user, server, storage, management, and out-of-band traffic as required. Assign ranges, gateways, and DHCP or static allocation rules. Publish the plan with the rack elevations.
Ensure operators can reach consoles, iLO/iDRAC, and network management when the production plane is impaired. Document jump-host or bastion expectations.
Define who has keys or badge access, visitor escort rules, camera or log expectations, and how access changes are approved and recorded.
Agree authentication for network devices and hosts, privileged account custody, baseline configs, and what must be disabled before go-live (default credentials, unused services).
Decide where authentication, config change, and critical system logs are sent, who reviews them, and how long they are kept for incident review.
Map demarcation, media converters, and who owns the circuit to the provider. Include escalation contacts and change notification expectations.
Define metrics and checks for PDU load, temperature, humidity if applicable, uplink health, critical hosts, and backup jobs. Prefer signals an on-call engineer can act on.
Map each alert class to owners, quiet hours policy, and escalation path. Silence without an owner is not an alerting strategy.
Require rack elevations, circuit maps, cable schedules, IP and VLAN plans, credential custody model, and vendor support contacts before acceptance is signed.
Cover power isolation, switch failover, cable replacement, and access recovery. Runbooks should match what was built, not a generic template.
Agree who approves rack adds, VLAN changes, and firmware updates while the environment is still stabilizing. Record the process operators must follow.
List critical spares (optics, PSUs, patch leads) and active support agreements. Note lead times that would block recovery if stock is empty.
Walk the room against the as-builts with facilities and ops. Capture deltas immediately; do not wait for the first incident to discover drift.
Assign named owners for power/facilities interface, network, compute, monitoring, and documentation updates. Shared ownership without a name is no ownership.
Services
Evidence
Industries
Resources
Next step
Share scope and constraints. We reply within 1–2 business days with fit and a practical approach.
AstraMakers