DEV Community

Dheeraj Ramasahayam
Dheeraj Ramasahayam

Posted on

DAC vs AOC vs Optic: Picking the Right Short Link in the Rack

Most server-to-switch links in a data center are short — a metre or two from the server NIC to the top-of-rack switch — and there are three ways to make them: a direct attach copper cable (DAC), an active optical cable (AOC), or a transceiver (optic) plus a fiber patch cord at each end. They all carry the same bits at the same speed, and they fail in completely different ways. Pick the wrong one and you pay for it twice: once in the price, and again at 2 a.m. when the link is down and the diagnostic you need does not exist on the thing you installed.

The three options, side by side

A DAC is twinax copper cable with the connectors built in — no separate transceivers. It typically reaches 1–3 m at 25G or 100G (up to 5–7 m at 10G), draws almost no power, costs the least, and reports no DOM readings because there is no laser to monitor. If one end fails, you replace the whole cable.

An AOC is fiber cable with transceivers built into both ends. It reaches tens of metres, is lighter and immune to electrical interference, and costs more than a DAC but less than two optics plus fiber at short range. Like a DAC it usually reports no DOM, and like a DAC it cannot be field-repaired — a fault at either end means replacing the whole cable.

An optic plus fiber patch cord is two pluggable transceivers and a separate fiber cord. It reaches hundreds of metres, costs the most for a short link, and is the only one of the three that reports DOM at both ends and lets you swap each part independently — the optic, or the cord, on its own.

DAC: the default for in-rack links

A passive DAC is copper wire with the connectors moulded on. No laser, no receiver, no DOM — the switch sees the link up or down and counts errors, and that is all the telemetry you get. That simplicity is the point: cheapest, lowest power, lowest latency. It is the right default for a server in the same rack as its switch, where the run is a metre or two.

Two things to check before you trust one. First, length: passive DAC reach shrinks as speed rises — plan on 3 m at 25G and 100G, longer only at 10G, and confirm against the cable datasheet rather than the label on the bag. An active DAC adds signal conditioning and stretches the reach to around 5 m, at a higher price and a little more power. Second, coding: many switches only accept DACs coded for their vendor, or a DAC on their compatibility list. A DAC that is electrically perfect and coded for the wrong vendor will sit there with no link and no error that explains why. Record the part number when you install one — it is the first thing the next tech needs when the spare has to match.

AOC: the middle distance

An AOC buys reach — tens of metres, enough to go from a server to a switch in the next rack or the end of the row — and a much lighter, thinner cable that is easier to route. What it does not buy is diagnostics or repairability. Most AOCs report no DOM, so a degrading link gives you no light-level warning before it fails, and you cannot test the fiber inside with a light source and power meter because the ends are fixed — there is no patch cord to unplug and measure.

Treat an AOC like fiber when you route it: respect the bend radius, never crush it under a cable manager lid, and never pull it by the connector. The cable looks like a DAC and gets handled like one; the fiber inside does not forgive that.

Optic plus fiber: the flexible one

Two pluggable transceivers and a fiber patch cord is the most expensive way to make a short link, and the most flexible. Each end is independently replaceable — a suspect optic moves to a known-good port to prove itself, a suspect patch cord swaps in seconds — and each optic reports DOM, so a failing laser shows up as falling Tx power or rising Rx loss before the link drops.

That diagnosability is why this is the right choice whenever the link matters enough to troubleshoot rather than just replace: cross-connects between racks, links that leave the rack, any run where you will be asked "is it the optic or the fiber?" The price of the flexibility is two more things to get wrong: the optic type has to match the speed and the fiber type (a single-mode optic on multimode fiber is a link that never comes up), and the optic has to be coded for the switch. Keep a known-good optic and a known-good patch cord in the spares kit — they are the test equipment for every other link in the rack.

The pick, in one paragraph

Same rack, under 3 m: passive DAC — cheapest, lowest power, and if it fails you just swap the cable. Between racks or to the end of the row, 3–30 m: AOC — cheaper than two optics, light to route, and you accept no DOM and no field repair. Longer than that, structured cabling through panels, or any link you expect to troubleshoot rather than replace: optic plus fiber — DOM at both ends and every part swappable. Mixed-vendor floor or a spares shelf that has to cover everything: optic plus fiber again, because one optic type and one patch cord type cover more combinations than a drawer full of coded DACs. If the link is already down and you are choosing under pressure, the safe choice is the one you can test: optic plus fiber, because it is the only one of the three that lets you isolate the fault instead of guessing.

When it is down: which fault belongs to which link

A DAC that never comes up is usually a coding or speed mismatch, not a broken cable: confirm both ports are set to the same speed, check the DAC is on the switch compatibility list or coded for that vendor, and try a known-good DAC of the same part number before you blame the port. A DAC that comes up and flaps under load is usually at its length limit or a damaged twinax pair — swap in a shorter DAC or move the server closer to the switch.

An AOC that never comes up is usually a bend or crush from installation, or the same coding problem as a DAC — inspect the whole run for a tight bend or a pinched point, then swap in a known-good AOC. An AOC that degrades slowly with rising errors and no DOM to read is a reason to replace the cable rather than chase the fault: there is no measurement inside it to take.

An optic link that never comes up is usually a type mismatch — check optic speed, single-mode versus multimode, and the fiber type in the patch cord, then read DOM on both ends: light present at the far end with no link means polarity or a config fault, no light at all means the near-end optic or the cord. Record which test you ran and what it proved; "swapped DAC, link up" is a diagnosis, "tried stuff, works now" is not.

Recording the link in the ticket

The part number is the record. A future tech cannot order or verify a link from "DAC cable" — they need the type, the length, the speed, and the part number, plus both endpoints and the verification. The minimum useful line:

srv-1042 Eth1 ↔ ToR SW-04 port 23,
100G passive DAC 2 m, part no. XYZ-100G-2M.
Link up, 0 errors in 10 min.
Replaced 3 m DAC (link flapped under load,
fault followed the cable).
Enter fullscreen mode Exit fullscreen mode

The shape to copy: endpoints, link type plus speed plus length plus part number, verification, and — if it was a replacement — what failed and what the swap proved. Five fields, and the next person on the ticket knows exactly what is in the rack and what to bring as a spare.

A question for you

When a short link goes down on your floor, what do you reach for first — a spare DAC, a spare AOC, or a known-good optic and patch cord — and what failure taught you to keep that one on the shelf?

Top comments (0)