If you’ve been following along with our 100G Ethernet series, you’ve run into a lot of vocabulary that gets used correctly but rarely explained: NRZ, PAM4, insertion loss, Type A versus Type B polarity, and a half-dozen acronyms that all seem to start with a letter that stands for “fiber.” This post is the answer key.
Nothing here is exotic. Most of it is old physics and older standards wearing new urgency, because the same conditions that a 10G link shrugged off for a decade will take down a 25G lane without warning. Bookmark this one — we’ll link back to it instead of re-explaining the same term in every post.
ASIC (Switch ASIC)
The dedicated chip inside a switch that actually moves traffic. Its design and the module you plug in decide which combinations of port speeds a switch can run at once.
BER (Bit Error Rate)
What fraction of bits arrive wrong, written in scientific shorthand: 1e-12 is one bad bit per trillion and healthy; 4e-5 is one per 25,000 and a lane about to become a ticket.
Cassette, Trunk, Patch Cord
The three pieces of a fiber channel. A trunk is the permanent multi-fiber cable running between rooms; a cassette is the panel-mounted box that breaks the trunk out into connectors; a patch cord is the short jumper from panel to equipment. Polarity has to be right across all three, and swapping one can break a channel that worked yesterday.
Clause 82 / Clause 91
Two sections of the IEEE 802.3 standard: one covers how a 100G stream is encoded and split across lanes, the other covers how its errors are corrected. You don’t need the internals — the operationally relevant fact is that on links using this encoding, FEC is a setting a human turns on at each end, and nothing in the protocol reconciles the two ends if they disagree.
CRC / FCS Errors and Discards
Three things your switch counts, and none of them mean quite what they sound like at 100G. A CRC error — which is what most switch CLIs print, though the more precise term is FCS (Frame Check Sequence), and an FCS is a type of CRC — means bit errors exceeded what FEC could correct, and the frame arrived bad. A discard means a frame was dropped on purpose, usually for congestion. FEC repairs damage before either counter is ever checked, so a lane can be bleeding errors quietly while both read zero.
CWDM (Coarse Wavelength Division Multiplexing)
Putting several signals on one fiber by giving each its own color of light. “Coarse” means the colors are spaced far enough apart to keep the optics cheap.
DAC (Direct Attach Copper)
A copper cable with a transceiver permanently attached at each end — no separate optic, no fiber required. Cheap and low-latency over short runs (typically a few meters), but heavier and less flexible than fiber, which is why you’ll find it inside a rack and rarely between racks.
FEC and RS-FEC
Forward Error Correction is extra data sent alongside the real data so the receiver can repair damaged bits itself, with no retransmission. RS-FEC is the specific Reed-Solomon scheme used at 25G and 100G, and it’s strong enough that a badly degraded lane gets repaired completely — which is exactly why the switch’s counters can read zero on a link that’s actually in trouble.
HCL (Hardware Compatibility List)
Your vendor’s official list of exactly which modules, cables, and configurations are supported together. Skipping it doesn’t always break something — but when it does, you’ll be troubleshooting alone instead of with a support contract behind you.
Insertion Loss
How much signal strength is lost end to end across a channel, measured in decibels. Every connector and splice costs a little; the question is whether the total stays inside the budget for the speed you’re running.
Margin
How much abuse a link can absorb and still work — the gap between the signal quality you have and the minimum you need. A 10G lane had enough margin to swallow a dirty connector without anyone noticing; a 25G lane doesn’t, so the same connector now causes errors. Margin is invisible on a healthy-looking link, which is the whole problem: it has to be measured rather than inferred.
MPO / MTP Connector
One connector housing many strands of fiber at once — typically 12, of which a 100GBASE-SR4 link only uses 8. MPO is the generic push-on interface; MTP is a trademarked, higher-performance version of it made by US Conec. It’s what makes parallel-fiber links practical, and it means one reseating mistake can affect several lanes at once.
Multimode and Single-Mode Fiber (MMF / SMF)
Multimode has a wide glass core and is used for short runs inside a building; single-mode has a narrow core and carries a signal much further. OM4, for example, is a multimode grade rated for 100G to about 100 meters — a rating on the glass, not on the current condition of your connectors.
NIC (Network Interface Card)
The network adapter inside a server. A 100G NIC only delivers 100G if everything behind it in the server — the PCIe slot included — can keep up.
NRZ (Non-Return-to-Zero)
The simplest way to send bits over light: on for a one, off for a zero, one bit at a time. Both 10G and 25G lanes use it, but the 10G lanes inside a 40G link run it with far more headroom than the 25G lanes inside a 100G link do.
OTDR (Optical Time-Domain Reflectometer)
An instrument that fires a pulse of light down a fiber and times what comes back, mapping every connector, splice, bend, and break by distance. It rules the cable plant in or out — necessary but not sufficient evidence that a link will actually carry 25G per lane.
PAM4 (Pulse Amplitude Modulation 4)
A way of encoding two bits per symbol instead of one, using four signal levels instead of two. It doubles the data rate over the same signaling rate as NRZ — which is how 25G electrical lanes become 50G and 100G optical links — but it also shrinks the margin between one signal level and the next. That shrinking margin is exactly why FEC becomes mandatory at these speeds instead of optional.
PCIe Gen3 x8
The server slot a network card sits in, described by generation and lane count. That combination tops out near 63 Gbps of real throughput — so a 100G NIC in one of these slots cannot actually reach 100G, no matter how healthy the network on the other end of the cable is.
Polarity, Type A / Type B
Polarity is making sure each transmitting strand lands on a receiving strand at the far end. Type A and Type B are standardized wiring methods that differ in how fiber positions are arranged inside the cable; neither is “correct” on its own, but the cassettes, trunks, and patch cords in a single channel all have to agree, or the link simply won’t come up.
Spine-Leaf
A network design where every access (leaf) switch connects to every core (spine) switch, and leaf switches don’t connect to each other. The leaf-to-spine uplinks are shared paths, so one degraded lane on one uplink can affect a lot of applications at once.
Strand
One individual fiber: a single thread of glass carrying light in one direction — except when using a BiDi (bidirectional) transceiver, which, despite what your instincts about light might tell you, sends signals in both directions over the very same strand at once. An SR4 link uses eight strands, and any single one of them can degrade without the link ever going down.
WDM / DWDM (Wavelength Division Multiplexing)
The general technique behind CWDM: stacking multiple signals on one fiber using different wavelengths of light. DWDM (Dense WDM) packs those wavelengths much closer together than CWDM does, which means more channels per fiber, pricier optics, and typically long-haul or carrier use rather than inside a data center.
Most of these terms only matter the moment something goes wrong — which is exactly when nobody wants to be looking them up for the first time. Keep this page handy.
- 100G Ethernet Explained: Standards, Speeds & Benefits
- QSFP28 Transceivers and 100G Ethernet
- MPO Breakout for 100G: Port Splitting, Configurations & Best Practices
- No Auto-MDIX Function on Fiber Interfaces – Time to Talk About “Polarity”
- Fiber Cable Types & Connectors
- Forward Error Correction (FEC) in 100G Networks
- Top 100G Ethernet Deployment Challenges & Solutions
- The Ins and Outs of Bidirectional Fiber Communication