Fiber is simpler than copper in one way: it is immune to electrical noise and does not care about grounding. It is less forgiving in another: a mismatched optic, polish, or polarity produces a dead link with no negotiation to explain it.
Two fibers
| Type | Core | Distance | Use |
|---|---|---|---|
| Multimode OM3 | 50 µm | 300 m at 10G | Building backbone |
| Multimode OM4 | 50 µm | 400 m at 10G | Building backbone, longer |
| Multimode OM5 | 50 µm | 400 m at 10G, wideband | New installs, SWDM |
| Single-mode OS2 | 9 µm | 10 km+ | Campus, ISP, long haul |
The rule of thumb: multimode inside and between nearby buildings, single-mode for anything that might grow. If a new campus is 2 km away, single-mode is the only answer.
Optics
| Form factor | Speed | Notes |
|---|---|---|
| SFP | 1 Gbps | Legacy access |
| SFP+ | 10 Gbps | Standard building uplink |
| SFP28 | 25 Gbps | New core |
| QSFP+ / QSFP28 | 40/100 Gbps | Core and aggregation |
A copper DAC is cheaper for short runs inside a rack; an AOC (active optical cable) is lighter and longer. Both bypass the patch panel, which is fine for a rack and wrong for a building.
The optic must match the fiber, not just the speed
A 10GBASE-SR optic is multimode at 850 nm. A 10GBASE-LR is single-mode at 1310 nm. Plugging an LR into multimode gives a link that never comes up, and plugging SR into single-mode gives a link that works only if you are very lucky and very short.
Connectors and polish
| Connector | Use | Notes |
|---|---|---|
| LC | Standard for SFP | Small form factor, duplex clip |
| SC | Older patch panels | Square, push-pull |
| ST | Legacy | Bayonet, being retired |
Polish matters: UPC (blue) and APC (green, angled) do not mate. An APC connector into a UPC port produces high loss and sometimes damage. Match the color, not just the shape.
Polarity
A duplex fiber pair has transmit and receive. Three polarity schemes exist (A, B, C), and a mismatch produces a link where each side transmits into the other's transmitter.
The practical rule: pick one scheme, document it, and use the same patch cords throughout. A single straight-through cord in a crossover scheme breaks one link and wastes an hour.
Worked example: building to building
Two buildings 380 m apart, 10 Gbps required.
| Option | Feasible | Notes |
|---|---|---|
| OM3 multimode | No | 300 m limit |
| OM4 multimode | Yes | 400 m limit, tight margin |
| OS2 single-mode | Yes | Plenty of margin, future 100G |
The OM4 option works today and is at the edge of its budget. Single-mode costs more per optic and removes the distance question permanently. For a link that will outlive two switch generations, single-mode is usually the better decision.
What breaks in production
- SR optic on single-mode (or the reverse), producing no link.
- APC into UPC, producing high loss or damaged ferrules.
- Polarity mismatch, where both ends transmit.
- Dirty connectors, the single most common cause of intermittent fiber faults.
- A bend tighter than the cable allows, producing loss that varies with temperature.
Checklist
- Fiber type and distance are documented for every run.
- Optic type matches the fiber (SR/OM, LR/OS2).
- Connector polish matches at both ends.
- One polarity scheme, documented and used consistently.
- Every connector is cleaned and inspected before mating.
- Spare optics and patch cords are stocked, matched to what is installed.
Next
Installation quality is invisible without measurement. Continue with Testing and Certification.