[ssl-3]

And the limitation of 1x 2.5G ethernet, for those who it actually makes a difference to, is broadly sidestepped by using a managed switch and VLANs (just like ye olde Linksys WRT54G -- a device that only really had a singular Ethernet interface -- did internally).

One can have as many 2.5G ports as one wishes -- just add more/bigger switches. They can be WAN or LAN or whatever you want. (They'll all share the same full-duplex 2.5G pipe to/from this router and the outside world beyond, but so what? OpenWRT makes it easy.)

[osti]

Is the idea to have modem and router on the same VLAN connected to that switch, and have the LAN traffic on another VLAN? So only one 2.5G port should be enough for everything?

[ssl-3]

Correct.

Switches vary in their nomenclature and defaults, but in the switch it could look something like this:

Port 1: Uber-fast WAN-device. VLAN 10 [WAN], untagged.

Port 2: Router-box. VLANs 1 [LAN, default] and 10 [WAN], tagged.

Ports 3-eleventy: Everything LAN. VLAN 1, untagged (or the default OOTB port config works for my stuff at home without any other considerations). More (cheap! dumb!) switches can be daisy chained here if one desires[1].

This keeps the traffic secure and separated. VLAN 1 [LAN] can't talk to VLAN 10 [WAN], and VLAN 10 can't talk to VLAN 1, but the router on port 2 can talk to both VLANs and it treats them as separate networks.

In OpenWRT, these individual VLANs show up almost like any other network interfaces do. The difference between a physical port and a VLAN member is pretty much just the name of that port. That part is easy -- in fact, it's all pretty easy.

---

And at the end of the day, this provides (up to[0]) ~2.5Gbps of WAN bandwidth to/from the LAN, minus whatever pittance of data is used for things like LAN DHCP or whatever other services are run on the router-box itself, in both directions, concurrently.

[0]: The most severe bottleneck here seems more likely to be the packet-processing capabilities of the dual-core 1.3GHz Cortex-A53 CPU than anything else. For instance: On the surface, it seems unlikely to have enough grunt to be able to do things like SQM at these kinds of data rates. But that bottleneck would exist with physical ports, too.

[1]: And in the context of home networking, daisy-chaining switches is not quite optimal, but it's also just fine, and it's far better and more-consistent than wifi. I just moved into a new place and I've already got 3 different network switches in use, for convenience as much as anything else, to get hardwired network access where it needs to be while running a minimum of cabling. (Maybe some day I'll go full structured cabling on the place with a rack of switches and shit somewhere, but meh: Just because I can doesn't mean that I should.)

[osti]

Thanks for the explanation!

I can only see one scenario where it doesn't work out to 2.5gbps, if we have symmetrical 2.5Gbps WAN, then to upload and download at 2.5g from a LAN device wouldn't reach 2.5, because the router has to use that single link for ingress and egress from both sources.

[ssl-3]

Ah. Yes, that's a thing. WAN <-> LAN bandwidth is multiplied by a factor of 2. I didn't work all the way out on the loop. (Apologies -- I don't work with constrained networks at this level every day, and I appreciate the correction.)

It may be fine for "normal" use: 2.5 in, or 2.5 out, or a mix of some of each. Unless we're heavy into torrenting, I'd like to suggest that most households aren't maxing out both at once: We're doing either a big upload or a big download (but usually not both at once), plus whatever baseload we have for background noise (streaming services, browsing, and whatnot).

It's good enough, maybe? Especially for asymmetric sources?

But really: If a person has needs symmetric 2.5, then maybe they should be shopping for a router-like-device that costs more than $90. This little box from the OpenWRT folks is pretty neat, but it's still certainly built down to a price. :)