MPO and MTP Fiber Polarity, Loss Budget and Testing

MPO is where most 40G/100G/400G turn-up problems come from, and almost none of them are the optics. They are polarity, loss budget or a dirty ferrule. This article covers the three things a data-centre engineer actually has to know: which polarity method your link uses, whether the loss budget closes, and what to measure before you declare the link good.

Polarity: A, B and C

MPO connectors have 8, 12, 16 or 24 fibers in a row. A duplex link uses two of those fibers (transmit and receive), so the cable has to map position 1 on one end to the correct position on the other. Three standards exist:

Method B is the most common in modern 40/100G deployments because it works with standard straight trunk cabling and 'flip' patch panels. The operational consequence is simple: never mix methods in the same link, and always note the method on the patch panel label. A-B mismatch produces a link that trains but reports errors, or never comes up at all, and it looks exactly like a bad optic.

Loss Budget Arithmetic

Allowed loss = Tx power (min) - Rx sensitivity (max)
Example, 100GBASE-LR4:
  Tx min    -4.3 dBm
  Rx max    +4.5 dBm  -> budget ~ 6.3 dB (per 100GBASE-LR4 spec, channel)

Consumed by:
  2 x MPO mated pairs (2 x 0.35 dB)     0.7 dB
  2 x LC mated pairs  (if patch)        0.4 dB
  Trunk 150 m @ 0.35 dB/km              0.05 dB
  Splice (if any)                       0.1 dB each
  Patch panels (2 x 0.2 dB)             0.4 dB
  Connector degradation allowance       1.0 dB
  ------------------------------------------------
  Total                                 ~2.7 dB  -> comfortable margin

Two rules of thumb that survive contact with reality: allow at least 3 dB of headroom for ageing, contamination and re-mating, and remember that single-mode MPO (especially 16/24-fiber for 400G) has tighter budgets than multimode. If your arithmetic leaves less than 3 dB of margin, the design is the problem, not the optic.

Connector Grades and Insertion Loss

MPO loss depends mainly on the ferrule grade: low-loss MPO is roughly 0.35 dB per mated pair, standard grade can be 0.6-0.75 dB. In a link with several MPO junctions, choosing standard-grade parts in a low-loss design is how budgets quietly disappear. If you are buying for 400G, specify low-loss and verify by measurement, not by the datasheet alone.

Testing Before Turn-Up

1. Inspect both ends. Contamination is the number-one cause of high loss, and it transfers to whatever you plug in next. Inspect and clean every connector before the first mate.
2. Measure with an OLTS (light source + power meter) or, better, a per-lane MPO test set - a single total-power reading hides one bad lane out of four or eight.
3. Record every lane. Per-lane results are what let you prove a later degradation is not your work.
4. Check polarity with an MPO tester or a known-good loopback before cabling the whole row.
5. Verify optics after patching: show interfaces Ethernet1 transceiver and the DOM values (...eeprom --dom on EOS) - receive power per lane confirms the physical layer as well as any meter.

Labeling and Documentation

An MPO install becomes unmaintainable without three facts on every label: the polarity method, the fiber count/type, and which lanes are in use for which link (for example lanes 1-4: leaf3-sw4 et1, 400G). Breakout cabling makes this mandatory - one 24-fiber trunk can carry six 40G or two 400G links, and somebody will re-patch it in a hurry during an incident.

Related reading: DAC, AOC and optics selection for when copper or active cables avoid MPO entirely, and the LR4/ER4/FR4 reach guide for picking the right optic class before you buy cable.

原文链接：https://www.thefoa.org/tech/Lossbudg.htm