What changes when the router is open hardware
A disaggregated router splits the software from the box. The network operating system runs on merchant silicon hardware that the buyer picks, and the port list becomes a purchase decision rather than part of the package. A white box switch is the same idea taken further: an open platform, a third party network operating system, and optics bought on their own merits.
On a closed platform the brand publishes the part list and the optics follow it. On open hardware there is no single list to follow: the cage, the network operating system release and the firmware on the box all have a say, and the compatibility matrix is something the operator holds rather than receives. This guide sets out what to fix before optics are quoted for an open platform. The column behind it is on the optical communication products page, and the rate families are collected in the optics selection guides.
Where the gate sits on an open platform
Three things decide whether a port comes up on open hardware. The first is the cage: a QSFP-DD cage and an OSFP cage are not interchangeable, and the form factor follows the chassis.
The second is the network operating system release. Some read the memory page and accept any part whose rate, media and lane count match the port profile; others check a vendor identity field, and a smaller group verifies a checksum that a blank page fails. The supported list moves with the release, so the same box can accept a part after a firmware step and reject it before.
The third is the port profile: speed, forward error correction mode, lane mapping and breakout come from the profile rather than the module label, so the profile name belongs on the requisition next to the part number.
Recommended optics for open routers and white box switches
| Module | Rate | Media / wavelength | Reach class | Where it usually lands |
|---|
| 1G and 10G SFP+, coded or uncoded | 1G / 10G | SMF or MMF, 850 / 1310 / 1550 nm | 300 m to 40 km class | Management, legacy and low rate access ports |
| 25G SFP28 | 25G | SMF or MMF, 850 / 1310 nm | 100 m to 10 km class | Server and access ports on open leaves |
| 40G QSFP+ and 100G QSFP28 | 40G / 100G | MMF parallel, or SMF CWDM / LAN-WDM | 100 m to 10 km class | Uplinks where the box is still 100G cabled |
| 200G QSFP56 and 400G QSFP-DD | 200G / 400G | MMF parallel or SMF DR4 / FR4 class | 100 m to 10 km class | Spine and storage ports on merchant silicon |
| 800G OSFP and QSFP-DD | 800G | SMF DR8 / 2xFR4 class | 500 m to 2 km class | New open fabrics feeding AI clusters |
| Breakout parts | Rate per lane | MPO parallel optics or SMF | Per lane class | One cage feeding four lower rate ports |
Reach classes are the standard ones for each family and the working limit is the measured loss on the route. Rate families are listed on the 10G, 25G SFP28, 40G QSFP+, 100G QSFP28, 400G and 800G pages.
Port speed, breakout and cage configuration
Merchant silicon ports are configured, not fixed. One 400G cage can run as a single 400G link, as four 100G links, or as two 200G links, and an 800G port can split into two 400G, four 200G or eight 100G positions. The split is decided in the profile, and the optics have to match it lane for lane.
Two failure modes come out of this. A breakout part that is not on the release's supported list is loaded but left down, and one on a port the profile still treats as a single wide port behaves the same way. Record the profile per cage, keep the lane mapping on the rack sheet, and treat a breakout position as a separate part line, because the part numbers differ.
Coding, network operating system releases and firmware
Open hardware does not mean optics without coding. The platform brand is no longer the deciding voice, but the release in use is, and a coded part removes the guesswork: the module is programmed for the platform named on the request, tested on the model, and it is not manufactured by that brand.
The practical requirement is a written matrix: box model, network operating system release, port profile, module part number and coding. Where an estate runs more than one release, ask for the part to be coded for the oldest release in production, since that is the unit that will refuse to come up. Coded parts for named platforms are collected on the Cisco, Juniper, Arista and Mellanox pages, and none of those parts is manufactured by the brand of the host.
Cooling, power and spares on open platforms
Open switches are usually bought on port cost and run dense, so check thermal headroom before the port count is fixed: confirm the airflow direction the chassis expects, the power available per cage and the temperature grade each position needs, then read the module diagnostics once the first links are up.
How to choose, in order
- List every box model with its network operating system release and firmware.
- Write the port profile per cage: speed, lane count, forward error correction mode and breakout.
- Measure each route including patch panels and set the reach class from the measurement.
- Fix the cable class per row and the connector type for the parallel links.
- Set the coding requirement per platform and confirm it against the release in service.
- Set the spare per module type and per breakout pattern, and keep the matrix with them.
Frequently asked questions
Do open platforms need coded optics at all?
Some releases accept uncoded parts on plain ports, and others check a vendor field or a checksum. State the box model and the release, and the coding is set to match.
Can one part number cover a closed platform and an open box?
Sometimes, when both read the same memory page and the open release does not verify a vendor field. It is a check on the real units, so we test the models named.
Will a breakout part work in every 400G cage?
Only where the profile and the release support it. A 400G to four 100G part needs the cage configured as four ports, and that setting is per port, not per box.
Is an 800G OSFP part usable in a QSFP-DD cage?
No. The form factors are mechanically and electrically different. Choose the form factor from the chassis first, then the rate and the reach.
What do you need to quote an open platform roll-out?
The box models with releases, the port profile per cage, the measured distances and loss, the cable class, the coding requirement, and the quantity per part with spares as separate lines.
Send the box list with releases, the profile per cage and the measured routes, and we come back with a coded part per position, the connector per link and the spare lines. Related reading: the 400G to 800G migration case, the 800G DR8, FR4 and ZR+ comparison, and all optics selection guides.
