TIA-942 spaces and topology

A large data centre holds tens of thousands of fibres and copper cables. Without a plan, every new server, switch and carrier circuit adds another cable run directly between two pieces of equipment, and within a few years the overhead trays are full of undocumented cable that nobody dares remove. Structured cabling replaces that tangle with a permanent, documented plant of standard links between defined spaces. Equipment then connects to that plant with short patch cords, and moves, adds and changes are made by repatching rather than by pulling new cable.

This lesson introduces the standards that govern data centre cabling, the functional spaces they define, the hierarchical star topology that connects them, and the choices between cross-connects and interconnects and between top-of-rack, end-of-row and middle-of-row layouts. The rest of the course fills in the details: media (Lesson 2), high-density fibre and loss budgets (Lesson 6), pathways (Lesson 8) and the outside plant that brings carriers into the building (Lesson 11).

The standards

Data centre cabling is designed to a small family of standards that work together:

• ANSI/TIA-942 (currently revision C, 2024), Telecommunications Infrastructure Standard for Data Centers: defines the spaces, topology, cabling and redundancy of a data centre, and a set of rating levels for the facility as a whole.
• ANSI/TIA-568 series: the generic cabling standards that set performance for balanced twisted-pair (568.2) and optical fibre (568.3) cabling and components, including test limits and polarity.
• ANSI/TIA-569: pathways and spaces, including conduits, trays, entrance facilities and the rooms that hold telecommunications equipment.
• ANSI/TIA-606: administration, meaning labelling and record keeping.
• ANSI/TIA-607: bonding and grounding of telecommunications systems.
• ANSI/BICSI 002, Data Center Design and Implementation Best Practices: a broad design reference covering architecture, electrical, mechanical and cabling, with its own availability classes.

In Canada, the CSA Group has published the TIA cabling standards as national standards (for example CSA T529 corresponds to TIA-568 and CSA T530 to TIA-569), and Canadian specifications commonly cite either the TIA or the CSA designation. The installation itself must also meet the Canadian Electrical Code, Section 56 for optical fibre cables and Section 60 for communication systems, plus the building code for fire ratings in plenums and shafts (Lesson 8).

Internationally, ISO/IEC 11801-5 and the European EN 50600 series cover the same ground with slightly different names: the "main distributor" and "zone distributor" correspond to TIA's main and horizontal distribution areas. Hyperscale operators often write their own standards that go beyond all of these, but the vocabulary of TIA-942 is the common language of the industry.

Functional spaces

TIA-942 divides a data centre into spaces by function. A small facility might combine several of them in one room, while a large campus may have dozens of each. The spaces are:

• Entrance room (ER): where carrier and campus cables enter the building and where carrier equipment (demarcation points, optical terminals) is located. It is the boundary between the outside plant and the building cabling. A colocation facility's "meet-me room", where customers cross-connect to carriers, is a form of entrance room. Large or highly available sites have two entrance rooms on opposite sides of the building.
• Main distribution area (MDA): the central point of the cabling system, holding the main cross-connect (MC) and usually the core routers and switches. Every other space is cabled back to it. Highly available designs add a second MDA.
• Intermediate distribution area (IDA): an optional second level, used in very large buildings or campuses, holding an intermediate cross-connect between the MDA and several HDAs.
• Horizontal distribution area (HDA): the distribution point for a group of rows, holding the horizontal cross-connect (HC) and often the LAN, SAN and management switches serving those rows. In a leaf-spine network, the HDA might hold the spine switches for a pod.
• Zone distribution area (ZDA): an optional consolidation point between the HDA and the equipment, such as a patch panel in an overhead enclosure or at the end of a row. It contains only passive cabling, never active equipment.
• Equipment distribution area (EDA): the racks and cabinets holding the end equipment, such as servers and storage, with the patch panels where horizontal cabling terminates.

Supporting spaces include the telecommunications room (TR), which serves offices and support areas outside the computer room, the operations centre, and the electrical and mechanical rooms, which have their own cabling for monitoring and control.

Topology: the hierarchical star

The spaces are connected in a hierarchical star. The MDA is the centre; backbone cabling runs from it to the entrance rooms, to any IDAs and to each HDA; horizontal cabling runs from each HDA (through a ZDA if one is used) to the EDAs it serves. There are no direct cables between HDAs or between EDAs in different groups in the basic topology, so any connection can be made by patching at the distribution points along a known path.

A star has three advantages for a cabling plant. First, every link has a predictable length and type, so its performance can be guaranteed in advance. Second, a change at one EDA affects only that EDA's horizontal links. Third, the cabling is independent of the network design: the same plant can support a three-tier network, a leaf-spine fabric or a storage network, because the logical topology is set by patching, not by pulling cable. TIA-942 allows direct backbone links between HDAs, or from an HDA to the ER, where a design needs them for redundancy or to bypass distance limits, but they are additions to the star, not replacements for it.

Backbone and horizontal cabling

The two levels of the star are treated differently:

• Horizontal cabling runs from the horizontal cross-connect to the outlets or patch panels in the EDA. For balanced twisted-pair copper, the permanent link (the fixed cable between two panels) is limited to 90 m, and the channel, including equipment and patch cords at both ends, to 100 m. Horizontal cabling is usually copper (category 6A) or multimode or single-mode fibre.
• Backbone cabling runs between the ER, MDA, IDAs and HDAs. Its length is limited by the applications it must support rather than by a fixed number, which in practice means optical fibre, mostly single-mode in new designs, with copper only for short links and for some carrier circuits.

Inside a data centre, the "horizontal" cabling is rarely horizontal in the building sense; it simply means the last tier of the star. Likewise, backbone cabling in a single-storey hall may run in the same overhead trays as horizontal cabling. The terms describe the cable's role in the topology, not its route.

Cross-connects and interconnects

At each distribution point, cabling can be connected to the equipment in two ways:

• Interconnect: the permanent cabling terminates on a patch panel, and a single patch cord runs from that panel to the switch port. Each link has one fewer connection, so it is cheaper and has less insertion loss.
• Cross-connect: the switch ports are permanently cabled to their own patch panel (an "equipment-side" panel), and the permanent cabling to the EDAs terminates on a separate panel. Changes are made only with patch cords between the two panels. The switch's cabling is never touched after installation, which protects dense line cards from repeated handling and lets changes be made in a dedicated patching area by staff who never open the network equipment cabinets.

The MDA normally uses cross-connects, because it holds the most valuable and most frequently changed connections. HDAs may use either. The extra connector in a cross-connect costs about 0.3 to 0.75 dB of optical loss and some copper performance margin, a cost that must be included in the loss budget (Lesson 6).

Equipment layouts: where the switches go

The structured cabling plan depends heavily on where the access switches (the first switch a server connects to, or the leaf in a leaf-spine fabric) are placed:

• Top-of-rack (ToR): a switch in every cabinet connects its servers with very short copper or direct-attach cables, and only the switch uplinks, usually fibre, leave the cabinet. This minimizes structured cabling to the EDAs, but needs many switches, some with unused ports, and one or two switch failures can isolate a whole cabinet.
• End-of-row (EoR): switches in a network cabinet at the end of each row connect to every server in that row through horizontal cabling. Fewer, larger switches are used more efficiently, but much more horizontal cabling is needed.
• Middle-of-row (MoR): like end-of-row with the network cabinet in the middle, halving the longest run.

Hyperscale and AI facilities almost always use ToR (or "top-of-rack plus spine row") designs because server racks arrive as fully cabled, pre-tested units and the facility cabling only needs to carry uplinks. Enterprise and colocation halls with mixed equipment more often use EoR or MoR with an HDA in each row, which looks like the classic TIA-942 star.

Redundancy and rating levels

TIA-942 rates data centres from Rated-1 (basic) to Rated-4 (fault tolerant) in four areas: telecommunications, architecture, electrical and mechanical. A facility's overall rating is the lowest of the four. For telecommunications the levels progress roughly as follows:

• Rated-1: a single entrance pathway and single paths throughout. Any failure or maintenance on a cable path can cause an outage.
• Rated-2: redundant active components (carrier equipment, routers, switches) but still single cable paths.
• Rated-3: two entrance pathways and diverse carrier routes into separate entrance rooms or at least separated points, redundant backbone pathways, and a second MDA recommended.
• Rated-4: fully redundant cabling with diverse backbone and horizontal pathways, a second MDA, and redundant HDAs, so no single failure of a pathway, room or component interrupts service.

BICSI 002 uses Classes F0 to F4, and the Uptime Institute's Tier I to IV classifications (covered in DCE-200) focus on power and cooling but expect the cabling to match. Whatever the scheme, the key ideas for cabling are the same: two physically separated paths for anything critical, and no single room whose loss isolates the facility.

Administration and labelling

A cabling plant is only useful if every link can be found. TIA-606 sets a scheme for labelling and records, organized around identifiers for spaces, cabinets, patch panels, ports and cables. In a data centre, cabinets are normally identified by a floor grid: if the computer room floor is laid out with a 600 mm grid, letters across one axis and numbers along the other give every tile a coordinate, and a cabinet takes the coordinate of the tile under its front right corner. A patch panel port might then be labelled "AJ05-B-24": cabinet AJ05, panel B (counting from the top), port 24.

Both ends of every cable carry the same label, which identifies both ends of the link, and the records (now usually a database or data centre infrastructure management, DCIM, system) list every link, its route, its test results and what it connects. Labels must be machine-printed and durable. Good administration is what allows abandoned cables to be identified and removed, and it is often the first thing an auditor or a new operator checks.

Designing a cabling plant: the process

A typical design proceeds in this order:

• Gather requirements: the network architecture, number of cabinets and switches, port counts and speeds, expected growth, redundancy level and the operator's own standards.
• Lay out the spaces: entrance rooms, MDAs and HDAs, located to keep runs short and pathways diverse.
• Choose the switch placement (ToR, EoR, MoR) and the cross-connect or interconnect strategy.
• Select media for each link from its length and speed (Lesson 2), and calculate loss budgets (Lesson 6).
• Count fibres and cables, plan trunks and patch panels, and allocate rack units.
• Design pathways: trays, routes, fill and separation (Lesson 8), and the entrance facilities and outside plant (Lesson 11).
• Write the labelling scheme and test requirements, and specify how the installer will document the as-built plant.

How this course is organized

The course moves from the overall design of a cabling plant to its parts and then back to the whole. Lesson 1 sets out the spaces and topology. Lessons 2 to 5 cover the media: choosing copper and fibre (Lesson 2), copper channel performance and Power over Ethernet (Lesson 3), the physics of optical fibre (Lesson 4) and the transceivers that set what the plant must carry (Lesson 5). Lessons 6 and 7 cover high-density fibre systems and loss budgets, then testing, certification and troubleshooting. Lessons 8 to 10 cover the physical plant: pathways and trays, cabling for AI clusters, and cabinets, cable management and administration. Lesson 11 takes the plant outside the building, and Lesson 12 brings everything together in a complete project from requirements to handover.