DAC vs AOC vs Fiber: SFP Compatibility Traps
The port is usually fine. The coding, speed, FEC mode, or fiber grade is not. A practical guide to DAC, AOC, and optical transceivers, with the commands that show what the module actually is.

On this page
- At a Glance: DAC, AOC, and Optical Transceivers
- What Is Actually in the Cage
- DAC, AOC, or Fiber: Pick by Distance and Power
- Trap One: The Port Is Faster Than the Optic
- Trap Two: The FEC Modes Do Not Match
- Trap Three: The Coding Says No
- Trap Four: The Fiber Is Wrong
- Trap Five: The Optical Budget Is Smaller Than the Run
- A Link Bring Up Procedure
- Record the Optic, Not Just the Port
A link that will not come up is rarely a mystery once you stop looking at the port and start looking at the module. The cage accepts the connector, the LED stays dark, and the switch log says the transceiver is unsupported or the link is down. Someone swaps the optic for another one from the same drawer and it works, which makes the first module look faulty. Often it was not faulty at all, just coded for a different speed, built for a different fiber, or set to a FEC mode the far end does not use.
The same confusion shows up in reverse when a link comes up but performs badly. The interface is up, the neighbor is reachable, and the error counters climb slowly enough that nobody notices until an application complains. That pattern is usually a physical layer problem: a dirty connector, a receive level outside the module’s window, or a FEC mismatch that the link tolerated just long enough to pass a ping test. This guide covers the three media people lump together as “the cable”: passive DAC, active optical cable, and separate optical transceivers. It then walks through the compatibility traps in the order they usually appear during a build, with the commands that show what the module actually is.
At a Glance: DAC, AOC, and Optical Transceivers
| Passive DAC | Active DAC | AOC | SR fiber optic | LR fiber optic | |
|---|---|---|---|---|---|
| Media | Twinaxial copper | Twinaxial copper with active ends | Fixed optical cable | 850 nm over multimode | 1310 nm over single mode |
| Typical reach | 3 to 5 m | 10 to 15 m | Up to 100 m | 100 to 400 m | 10 km |
| Power per end | Near zero in the cable | Roughly 0.5 to 1 W | Roughly 1 W | Roughly 1 W | Roughly 1 to 1.5 W |
| Connector at the cage | Fixed | Fixed | Fixed | SFP or SFP+ cage | SFP or SFP+ cage |
| Replaceable ends | No, replace the cable | No, replace the cable | No, replace the cable | Yes, replace one optic | Yes, replace one optic |
| Best fit | In rack server and switch links | Short links that need more reach | Rack to rack within a room | Data center short reach | Building, campus, and carrier handoff |
| Main failure mode | Cable damage or bend | Cable damage or active end failure | Cable damage or active end failure | Dirty connector or wrong grade | Loss budget or connector polish |
What Is Actually in the Cage
The cage is a mechanical and electrical interface, not a promise about what the module can do. An SFP cage is commonly 1G, an SFP+ cage 10G, an SFP28 cage 25G, a QSFP+ cage 40G, and a QSFP28 cage 100G. A QSFP-DD cage is 400G. The module inside determines the media, the wavelength, the reach, and often the coding that the host will accept.
Every optical module and every DAC or AOC end carries an EEPROM with identification data. The host reads that data to decide whether the module is supported, what speed it can run, and whether digital optical monitoring is available. That EEPROM is why two modules that look identical can behave differently in the same port.
Start by reading what the switch sees. On Cisco IOS the compact view gives temperature, voltage, bias current, and the two optical power readings.
show interfaces transceiver
Optical Optical
Temperature Voltage Current Tx Power Rx Power
Port (Celsius) (Volts) (mA) (dBm) (dBm)
--------- ----------- ------- -------- -------- --------
Te1/1/1 38.1 3.29 6.7 -2.9 -3.4
The detail variant adds the alarm and warning thresholds, which is what you actually need when deciding whether a reading is normal for that module.
show interfaces Te1/1/1 transceiver detail
Junos presents the same data under a different command.
show interfaces diagnostics optics xe-0/0/0
Laser bias current : 6.744 mA
Laser output power : 0.5520 mW / -2.58 dBm
Receiver signal average optical power : 0.0000 mW / - Inf dBm
A receive reading of - Inf dBm is not a low signal. It means the module is not seeing light at all, which points to a missing fiber, a broken strand, or a transmitter that is shut down. That distinction saves time before anyone starts cleaning connectors.
On a Linux host, ethtool reads the same EEPROM.
ethtool -m eth0
Identifier : 0x03 (SFP)
Connector : 0x07 (LC)
Transceiver type : 10G Ethernet: 10G Base-LR
Laser wavelength : 1310nm
Length (SMF,km) : 10km
Vendor name : FS
Vendor PN : SFP-10G-LR
Optical diagnostics support : Yes
Laser output power : 0.5012 mW / -3.00 dBm
Receiver signal average optical power : 0.3981 mW / -4.00 dBm
Module temperature : 38.00 degrees C / 100.40 degrees F
The Transceiver type, Laser wavelength, and Length fields are the ones that tell you whether the module matches the fiber. The vendor and part number tell you what to order when it fails.
DAC, AOC, or Fiber: Pick by Distance and Power
The decision is mostly distance, then power, then whether you need to replace one end independently. A passive DAC is the cheapest and lowest power option for a link inside a rack, and it has no optics to clean or code. Its limit is electrical: the signal degrades quickly over copper, and the reach shrinks as speed rises.
An active DAC adds a small active circuit at each end to extend the copper reach. An AOC goes further by converting the signal to light inside the cable, which makes it thinner, lighter, and immune to the electrical limits that cap DAC. Both are fixed assemblies, so a damaged end means replacing the whole cable.
Fiber separates the cable from the optics. That costs more, consumes more power, and adds two connectors that can be dirty, but it is the only option for long runs and the only one that lets you replace a failed optic without touching the cabling.
Power is the second axis. A 10GBASE-SR module typically draws around 1 W, a 10GBASE-LR around 1 to 1.5 W, and an AOC is in the same range per end. A passive DAC has almost no active electronics in the cable, but the host SerDes still burns power to drive it. In a dense rack the difference between a DAC and an LR optic across 48 ports is a real thermal number, not a rounding error.
The practical rule is simple. Use passive DAC inside the rack, active DAC or AOC for the short run to the next rack, SR fiber for the data center reach that exceeds copper, and LR or ER fiber for anything that leaves the room. If the link might grow beyond the current distance, fiber is the safer investment because the cable can stay while the optics change.
Trap One: The Port Is Faster Than the Optic
A cage that fits is not a port that supports the module. The common versions are SFP at 1G, SFP+ at 10G, SFP28 at 25G, QSFP+ at 40G, QSFP28 at 100G, and QSFP-DD at 400G. A 10G SFP+ module can sometimes run in an SFP28 cage at 10G, because the cage is backward compatible and the port supports the lower speed. A 25G module in a 10G only cage cannot, because the port has no 25G signaling to negotiate.
The same logic applies to 1G. Some SFP+ ports accept a 1G SFP and run at 1G, and some do not. The datasheet and the port capability list are the authority, not the shape of the hole. show interfaces Te1/1/1 reports the negotiated speed and the media type the switch believes is installed, which is enough to spot a mismatch before you replace anything.
If the link is down, the media type still tells you what the switch believes is installed. If it reports a 10G module and the far end is a 25G only port, the mismatch is already visible.
Breakout cables add another layer. A 40G QSFP+ port can sometimes be split into four 10G links, and a 100G QSFP28 port into four 25G links, but only if the platform supports breakout on that port group and the correct breakout cable or harness is used. The port group configuration is platform specific, so check the switch model’s breakout guide rather than assuming the cable alone enables it.
Trap Two: The FEC Modes Do Not Match
Forward error correction is the trap that looks least like a cabling problem. At 10G it is often absent or optional. At 25G and above it is common, and some optics require a specific mode. If one end expects RS-FEC and the other expects no FEC or Fire Code FEC, the link may stay down, flap, or come up with a rising error count.
Cisco Catalyst platforms expose the setting per interface.
interface TwentyFiveGigE1/1/2
fec auto
show interfaces TwentyFiveGigE1/1/2 | include Fec
Fec is auto
Junos uses clause numbers in the interface configuration.
set interfaces et-0/0/17 gigether-options fec fec108
The fec108 value is the IEEE 802.3by Clause 108 Reed-Solomon mode used on 25G links. fec74 is the older Fire Code mode, and fec91 is the Clause 91 mode used at 100G. The names matter because a mismatch between fec74 and fec108 is not a negotiation failure you can see in the link state alone.
On Linux, ethtool --show-fec eth0 reports the configured and active mode, for example Configured FEC encodings: Auto and Active FEC encoding: RS.
The symptom to look for on a link that is up but unhealthy is a FEC error counter that climbs while the interface stays up.
show interfaces xe-0/0/0 extensive | match FEC
Ethernet FEC Mode : FEC108
FEC Corrected Errors : 148223
FEC Uncorrected Errors: 37
Corrected errors are not automatically a failure, but a rapidly rising corrected count means the margin is thin. Uncorrected errors mean frames are being lost. Either way, the next step is to compare the FEC mode on both ends before replacing the optic.
Trap Three: The Coding Says No
The EEPROM coding is the part of the module you cannot see by looking at it. Vendors encode the module type, the speed, the vendor name, and a checksum. Some switches check that data strictly and refuse anything that is not on their list, even when the module is electrically fine.
The log entry is usually explicit.
%PHY-4-UNSUPPORTED_TRANSCEIVER: Unsupported transceiver found in Te1/1/1
%PM-4-ERR_DISABLE: gbic-invalid error detected on Te1/1/1, putting Te1/1/1 in err-disable state
Cisco platforms have a long standing pair of hidden commands that relax the check.
configure terminal
service unsupported-transceiver
no errdisable detect cause gbic-invalid
end
The first command tells the switch to accept modules it would otherwise reject. The second stops the port from being error disabled for the specific gbic-invalid cause. Both are deliberate operational choices with support implications, and some newer platforms do not implement them at all. Treat them as a temporary measure with a known risk, not as a standard build step.
The better long term answer is to standardize on a small set of coded modules per platform and keep the part numbers in the inventory. A module that is “the same speed” but coded for a different vendor or a different reach is exactly the kind of spare that works in one switch and fails in another.
Trap Four: The Fiber Is Wrong
The optical module and the fiber have to agree on mode and wavelength. A short reach module is built for 850 nm over multimode fiber. A long reach module is built for 1310 nm over single mode. An extended reach module uses 1550 nm over single mode. Plugging a single mode optic into multimode fiber, or the reverse, usually produces no link at all, and occasionally produces a link that works just well enough to be confusing.
| Module | Wavelength | Fiber | Typical reach |
|---|---|---|---|
| 10GBASE-SR | 850 nm | OM3 or OM4 multimode | 300 m on OM3, 400 m on OM4 |
| 10GBASE-LR | 1310 nm | OS2 single mode | 10 km |
| 10GBASE-ER | 1550 nm | OS2 single mode | 40 km |
| 25GBASE-SR | 850 nm | OM4 multimode | 100 m |
| 100GBASE-SR4 | 850 nm | OM4 multimode | 100 m |
| 100GBASE-LR4 | 1310 nm | OS2 single mode | 10 km |
Connectors add their own traps. LC is the usual duplex connector for SFP and SFP+ optics. MPO or MTP is used for parallel optics such as 40GBASE-SR4 and 100GBASE-SR4, and the polarity of an MPO trunk matters. A Type A, Type B, or Type C polarity is not interchangeable, and a swapped pair in a duplex LC link produces the same no light symptom as a broken strand.
Polish matters too. UPC and APC end faces are not compatible. An APC connector has an angled ferrule and a UPC connector has a flat one, and mating them can damage both surfaces. Most data center optics use UPC. Telco and long haul plant often use APC. Match the polish at every patch panel and every adapter.
Trap Five: The Optical Budget Is Smaller Than the Run
A fiber link has a loss budget, and the installation consumes it. The budget is the difference between the transmitter’s minimum output and the receiver’s sensitivity. Every kilometer of fiber, every connector pair, and every splice takes a slice of that budget. When the slices add up to more than the budget, the link either fails or runs with almost no margin.
A 10GBASE-LR link is a useful worked example. The transmitter minimum is -8.2 dBm and the receiver sensitivity is -14.4 dBm, which gives a budget of 6.2 dB.
| Item | Loss | Unit | Why it matters |
|---|---|---|---|
| Transmit budget | 6.2 | dB | Minimum transmit power minus receiver sensitivity |
| 10 km of OS2 at 1310 nm | 3.5 | dB | 0.35 dB per kilometer |
| Two LC connector pairs | 1.0 | dB | 0.5 dB per pair, worse if dirty |
| Four splices | 0.4 | dB | 0.1 dB per splice |
| Remaining margin | 1.3 | dB | Budget minus the losses above |
The reverse problem is just as real. A long reach optic on a short patch cord can overload the receiver. If the receive power is above the module’s maximum, the link may work intermittently or not at all. The fix is an inline attenuator sized to bring the receive level into range, not a different cable.
The DOM readings are the evidence. Compare the reported receive power with the datasheet window, then compare it with what the budget predicted. A reading that is 3 dB lower than expected usually means a bad connector, a damaged strand, or an extra patch panel nobody documented.
A Link Bring Up Procedure
The order matters because each step removes a class of causes. Do them in this sequence and you will not spend an hour replacing an optic that was never the problem.
- Confirm the port speed and form factor. The module must match what the cage supports, and the port must be configured for the right speed.
- Confirm the module media matches the fiber. Check the wavelength, the mode, and the connector type before cleaning anything.
- Confirm the FEC mode on both ends. If the link is 25G or 100G, this is not optional.
- Read the DOM values. Transmit and receive power should both sit inside the module’s window, not merely inside the switch’s alarm thresholds.
- Check the error counters. A link that is up with CRC errors is not a working link.
- Record the module and the path. The next person needs the part number, the serial number, and the fiber pair, not just the port name.
The error check is quick on Cisco.
show interfaces Te1/1/1 | include "input errors|CRC|output errors"
0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored
0 output errors, 0 collisions, 0 interface resets
On Junos, the extensive output includes the FEC counters and the media state.
show interfaces xe-0/0/0 extensive | match "errors|CRC|FEC"
On Linux, the driver decides the counter names, so list them and filter.
ethtool -S eth0 | grep -Ei 'err|drop|crc|fec'
The exact names vary by driver. The point is to look at the counters that represent physical errors, not just the ones the vendor put first.
Record the Optic, Not Just the Port
A port label tells you where the cable goes. It does not tell you what module is in the cage, what reach it has, or whether the spare in the drawer will work. Record the form factor, vendor, part number, serial number, wavelength, reach, connector, and the fiber pair or patch panel ports on each end.
That record pays off twice. When a module fails, the part number is already known. When a link is upgraded, the existing fiber grade and connector polish are already known, so the new optic can be chosen without sending someone to trace a cable.
The transceiver form factor reference is a quick way to check the speed and reach of a module family. For the physical path, cable tracking keeps the endpoints and the cable record together, which is where most link bring up time is actually lost.
The first step is small. Pick one rack, run the transceiver command on every port, and write down the part number and DOM reading for each link. The gaps in that list are the links that will cost you the most time when they fail.
Frequently Asked Questions
What is the difference between DAC, AOC, and fiber optics?
How far can a passive DAC cable reach?
Why does a compatible SFP show up as unsupported?
Can I plug a 10G SFP+ into a 25G SFP28 port?
What is FEC and why does it break links?
How do I read the DOM values on a transceiver?
How should I document optics and cables?
Stop reaching for a spreadsheet
Obelinf keeps every subnet, device, circuit, and rack in one live source of truth, with audit logs and a topology view. Free for personal use.
Related Articles

VRRP vs HSRP vs GLBP: First Hop Redundancy
Every host has one default gateway, and that gateway is a single point of failure until you make it virtual. Compare VRRP, HSRP, and GLBP with real configs, timers, and failure checks.
Read more
Fiber Optic Cable Types Explained: OS1, OS2, OM3, OM4
Understand single mode and multimode fiber grades, what OS1, OS2, OM3, OM4, and OM5 mean, how far each reaches, and how to pick the right cable for a run.
Read more
What Is a CDU? Facility Water, In Row, and In Rack Explained
A practical guide to coolant distribution units: the two loops they separate, the three form factors, how approach temperature sets the plant, and how to size flow from rack heat load.
Read more