transceiver-db/blog-training-data/blog-004-400g-migration-fiber-plant.md

title	type	audience	quality_score	generated_by	generated_at	training_data
Your 100G Fiber Plant Is Not Ready for 400G. Here's How to Find Out Before It Bites You.	tutorial	network_engineers_dc_operators	9	claude-sonnet-4-20250514	2026-04-06	true

The link won't come up. Or it comes up, holds for three minutes, then drops. Or it's up but BER is drifting and you can't figure out why. You've replaced the optic twice. You've swapped the cable. The switch vendor TAC is asking for logs you've already sent them.

There's a good chance the problem is your fiber plant.

Specifically: cabling infrastructure that worked fine for 100G SR4 or even 100G LR4 has a meaningful probability of being marginal for 400G DR4 — not because anything broke, but because the loss budget at 400G is tighter and your plant was never characterized to the margin it now requires.

Here's what changes at 400G.

QSFP28 100G SR4 over OM4 has a maximum reach of 100m and a total optical budget of around 7.6 dB. That's generous. A slightly dirty connector, a patch cord with 0.5 dB insertion loss instead of 0.3, a couple of aging splice closures — the budget absorbs it. 400G QSFP-DD DR4 over OS2 singlemode has 500m of reach, which sounds like more room, but the available link budget for the entire span, including connectors and splices, is approximately 6.5 dB. That's the entire budget. No forgiveness. A single dirty end-face that would have been invisible at 100G can cost 1-2 dB on a contaminated MPO-12 interface, and now you're at margin. Maybe below it.

The failure mode isn't always dramatic. Sometimes you get no link. More often — and more insidiously — you get a link that functions at BER levels that are just below the FEC correction threshold under normal conditions, but tips over that threshold under thermal load, traffic bursts, or minor physical perturbation (someone brushes the cable tray, a fiber moves by a millimeter). Post-FEC errors start climbing. You get traffic drops that don't correlate to anything visible in syslog. This is the 400G deployment failure pattern that's hardest to debug, because it doesn't fail cleanly.

The diagnostic path starts at the MPO-12 interface.

Pull the fiber. Inspect the end-face with a fiber inspection probe — a visual inspection tool, not a power meter. What you're looking for is contamination, scratches, or chips in the core. Every MPO-12 connector has 12 fibers in a single interface. One contaminated fiber in that array degrades one lane. DR4 uses four transmit and four receive lanes. If any of those lanes is compromised, you have a partial link failure that presents as an asymmetric BER condition across the four lanes.

Clean it. This matters more than it sounds. A standard MPO cleaning tool (dry cleaning cassette, lint-free swab with IPA, or air clean depending on what you have) removes contamination that genuinely costs 1-2 dB. If you haven't cleaned the connectors recently, do it before you do anything else in the diagnostic chain. The number of 400G failures that resolve with end-face cleaning is high enough that cleaning is step one, every time, no exceptions.

After inspection and cleaning, take a loss measurement. You need an optical power meter or an OTDR, not the DOM Rx power reading from the switch CLI. The DOM reading tells you what power is arriving at the photodetector — it's useful but it doesn't break down the loss sources. An OTDR trace shows you loss by distance: you can see splice events, connector events, and whether a specific location in the span is introducing unexpected loss. For a new 400G deployment or a troublesome existing one, an OTDR trace on each fiber in the MPO is worth the time it takes.

The numbers to hold in your head for 400G DR4 on OS2 singlemode: 0.35 dB/km fiber loss at 1310nm (DR4 operates at 1310nm, not 1550nm — this is a common mistake and the loss figures are different), 0.3 dB per connector under clean conditions, 0.1 dB per fusion splice, 3 dB margin minimum. Run the budget with those numbers for your actual span. If the theoretical loss plus margin exceeds 6.5 dB, you have a margin problem that no transceiver replacement will fix.

The fiber type question catches some teams by surprise. If the cabling infrastructure was installed during a 10G or early 100G era, there may be OM3 or OM4 multimode fiber in the plant. DR4 requires OS2 singlemode. SR4 requires multimode. These are not interchangeable. Putting a DR4 transceiver on a multimode cable doesn't give you a link that degrades gracefully — it gives you nothing, or at best extremely high BER because the modal characteristics of multimode fiber at 1310nm with a singlemode source produce unusable output. If you're inheriting an infrastructure build and don't have a fiber plant documentation, pull the spec sheet for the installed cable before you spec the optics.

One pattern that appears repeatedly in 100G-to-400G transitions: the existing plant uses short MPO trunk cables with LC breakouts at the patch panels. That works well for SR4 (which is also MPO, also 8-fiber, also multimode). The same physical plant with OS2 trunk cables should work for DR4 — but the breakout loss at the cassette matters more than it did before. Verify the insertion loss specification on the cassette itself, not just the trunk cable. Some cassette designs introduce more connector pairs than others. Every connector pair is another 0.6 dB worst-case.

The good news: a fiber plant that's causing 400G failures is usually fixable without replacing cable. End-face cleaning, cleaning cassette replacement, occasionally a bad patchcord swap — these resolve the majority of cases. What they require is doing the characterization work before deployment rather than after the first outage.

Running a power budget calculation before installation takes ten minutes. Running it from a production switch while traffic is impacted takes considerably longer and costs considerably more.