DEV Community

Dheeraj Ramasahayam
Dheeraj Ramasahayam

Posted on

Single-mode vs multimode: picking the right fiber for the run

Single-mode vs multimode: picking the right fiber for the run

The single-mode vs multimode decision is made once, at pull time, and it is expensive to undo. Pull the wrong one and the fix is not a different optic — it is a new cable, a new ticket, and a new maintenance window, because the fiber type is the span. The decision itself is simple once you stop comparing the fiber and start comparing the link: how far, how fast, and how long this run has to last. Answer those three and the fiber picks itself.

What each one actually is

Single-mode fiber (OS2) has a ~9-micron core and carries one light path, usually at 1310 or 1550 nm. One path means almost no modal dispersion, so the signal travels kilometres without smearing — a 10G LR optic reaches 10 km over OS2, and a 100G LR4 optic reaches 10 km too.

Multimode fiber (OM3, OM4, OM5) has a 50-micron core and carries many light paths at 850 nm. Many paths means the signal spreads as it travels, so the reach is limited by speed: the faster you go, the shorter the run.

The jacket tells you what you are holding — single-mode OS2 is yellow, OM3 and OM4 are aqua, OM5 is lime green, and legacy OM1/OM2 is orange — but the jacket is a clue, not proof. Read the print on the cable and the rating on the span record before you trust it.

Distance and speed: the table that decides it

These are the standard reaches for short-reach (SR) multimode optics and long-reach (LR) single-mode optics. If your run is longer than the number in the table, that fiber at that speed is not an option — no optic swap fixes it.

Speed / optic OM3 OM4 OM5 OS2 (single-mode)
10G SR / LR 300 m 400 m 400 m 10 km (LR)
25G SR / LR 70 m 100 m 100 m 10 km (LR)
40G SR4 / LR4 100 m 150 m 150 m 10 km (LR4)
100G SR4 / LR4 70 m 100 m 100 m 10 km (LR4)
400G SR8 / DR4 — 100 m 150 m 500 m (DR4)

Two things jump out. First, multimode reach collapses as speed rises: a 400 m OM4 run that is fine at 10G is unusable at 100G. Second, single-mode reach barely moves with speed — the same OS2 that does 10 km at 10G does 10 km at 100G and 500 m at 400G. That is the whole argument for single-mode on anything that has to last.

The real cost comparison

The common version of this decision — "multimode is cheaper" — is true and misleading at the same time. The fiber itself is usually cheaper per metre for single-mode; the optics are where the money is. A single-mode LR optic costs several times a multimode SR optic at the same speed, so a short multimode link is genuinely cheaper overall: cheap fiber, cheap optics, done.

But the comparison flips the moment the run has to change. A multimode run that cannot reach the next speed has to be re-pulled — new cable, new tray work, new window, new ticket — and a re-pull costs far more than the optic premium ever did. Price the link over the life of the run, not over the day it is installed.

When single-mode wins even on a short run

Pull single-mode when the run is a backbone, a trunk between panels, a campus link, or anything you expect to outlive the current speed. A trunk pulled once in OS2 never has to be re-pulled for a speed upgrade: 10G today, 100G tomorrow, 400G after that, same fiber, new optics at the ends.

That is why data centers standardize on single-mode for trunks even when the distance is only tens of metres — the fiber is the permanent part of the plant, and the optics are the part that changes. If you are pulling fiber through a tray you never want to open again, pull single-mode.

When multimode is the right call

Pull multimode for short, fixed-speed patch links where the saving is real and the speed is settled: in-rack and intra-row links, server-to-ToR at 10G or 25G, short runs inside one row where the distance is metres and the optics are SR. The SR optics are cheap, the fiber is easy to terminate, and the link will never need to go faster than the table allows because the endpoints are fixed.

The test is simple: if you can name the speed this link will run for its whole life, and the distance fits the table with margin, multimode is the cheaper correct answer. If either answer is "we don't know yet," that is a single-mode run.

The trunk rule, and the one thing never to do

The rule I use on the floor: backbone and trunks in single-mode, short fixed-speed patch in multimode, and never mix the two on one span. A span is one fiber type end to end — single-mode optic to single-mode fiber to single-mode optic, or multimode to multimode. Mixing them is not a degraded link, it is a link that never comes up or flaps with errors, because the core sizes and wavelengths are matched sets.

If you inherit a link that never comes up, check the three facts before swapping anything: jacket color, optic type (SR/SX means multimode at 850 nm; LR/LX/ER means single-mode at 1310 or 1550 nm), and the span record. If any one disagrees with the other two, you have found the mismatch.

Recording the choice in the ticket

The fiber type is part of the link record, so it belongs in the ticket. The minimum useful line:

Panel A tray 2 ↔ Panel B tray 4, OS2 single-mode, yellow, 45 m trunk, 10G-LR optics both ends, 1310 nm. Span loss 0.8 dB vs 2.5 dB budget — pass.

Six fields: endpoints, fiber type and rating, jacket color, length, optic pair and wavelength, and the measured loss against the budget. The next tech on the ticket knows what fiber is in the tray, what optics it takes, and whether the span has headroom for the next speed — which is exactly the information the person who has to upgrade it will need.

A question for you

What is the longest run in your plant that you would still pull in multimode — and what made you draw the line there?

Top comments (0)