How do you design structured cabling for a data centre?

A common way to get data centre cabling wrong is to start with the cable: pick a category or a fibre type and work backwards. That leaves the important questions unanswered.

Structured cabling is the permanent, organised cabling between equipment cabinets, distribution points and external connections, installed to a standard and documented so that connections can be added and changed by patching, without pulling new cable. In a data centre it should be designed from the network architecture outwards: first which devices connect to which, at what speeds and over what distances; then the topology of distribution points; and only then the physical pathways.

The standards

Two families of standards are commonly used. ANSI/TIA-942, now in revision C, covers the telecommunications infrastructure of data centres and computer rooms, as well as site, architectural, electrical, mechanical, fire safety and security aspects. In the international ISO/IEC 11801 series, ISO/IEC 11801-5 specifies generic cabling within and to the computer-room spaces of data centres, while office and industrial spaces in the same premises fall under Parts 2 and 3. The designer should state which standard the design follows, because mixing them without saying so creates ambiguity at testing and acceptance.

Designing from the network outwards

The design begins with a list of the networks the facility will carry: the production data network, storage, out-of-band management, any high-performance compute fabric, building and security systems, and carrier connections. For each, it records port counts, speeds and the interface types the equipment supports.

Equipment distance limits come next, and high-performance fabrics impose constraints that ordinary enterprise networking does not. GPU cluster designs, for example, can set a maximum total cable path for every fabric link, and the vendor's documentation for the exact system is the place to find it. If a power limit then forces racks to be spread across more floor space, cable lengths must be recalculated and latency may be affected. Cabling, power and layout are a single design problem.

With the networks and limits known, the topology can be drawn: where the main and zone distribution points sit, how cabinets connect to them and how carriers enter the building, with room left for the networks that are known to grow. The pathways follow, whether overhead trays or under-floor routes, with separation between power and data, fire-stopping and physical diversity where resilience matters.

The pathways must also be coordinated with cooling. The US Department of Energy's data centre design guide warns that cable congestion in raised-floor plenums can sharply reduce airflow and promote hot spots. It recommends a cable management strategy for the whole air path, a programme to remove abandoned cables, and sealing of cable openings in floors and ceilings.

Last comes the information that keeps the installation usable: consistent identifiers for cabinets, panels, ports and links, a test regime for every installed link against the chosen standard, and a process that keeps the records current after moves and changes.

Sizing for growth

Spare capacity should follow a growth scenario, not a blanket percentage. What is the next stage of the network likely to be? Which pathways and panels would it use, and could it be installed without disturbing live services? Reserving tray space and panel positions is usually cheaper than reserving cables that may be the wrong type by the time they are needed.

Who decides what

Cabling sits between several disciplines, and gaps open up wherever each assumes another has made the decision.

What the tender should say

• The standard and edition the design follows, and the performance class or category for each link type.
• The test method and pass criteria for each link type, and the form in which results are delivered.
• The labelling scheme and the record format, including how records load into the operator's inventory.
• The growth scenario that spare pathway and panel capacity must serve.
• Any manufacturer system warranty you require, and the installer qualifications its conditions demand.
• How variations are priced and recorded, so late changes still appear in the as-built records.

Accepting the installation

Acceptance is where the earlier decisions are proved. Every link should have test results against the specified standard and performance class, and the labels at both ends should match the records. Installed routes, including the diverse ones, should match the drawings, and penetrations should be properly fire-stopped and sealed. The as-built records should arrive in a form the operator will actually maintain, ideally linked to the DCIM inventory.