The h/w components of my network all seem to be reaching EoL at around the same time, so I've replaced my router with a new GL iNet MT6000, and my old unmanaged switches with 3 second hand Zyxel GS1900-24HP switches for a "drop in upgrade" to give me managed networking and PoE. My wiring is only cat5e, so I'm still limited to 1Gbps.
And that was fine, but this week I noticed that one of my trusty old BT Homehub 5A access points is starting to struggle too (random lock-ups). I'm impressed with the MT6000 router, so I'm tempted to replace my three access points over the next couple of months with new GL iNet AX3000 "Beryl AX" devices. I figure that will give me enough future proofing for the next few years at least.
So, my question: the Beryl AXs are theoretically capable of ~3Gbps of WiFi throughput, and come with a 2.5Gbps uplink, as well as a 1Gbps LAN port. With the WiFi now able to move so much data, and my wired network only capable of 1Gbps, is it worth implementing Link Aggregation to give me more capacity (not speed!)?
So linking the two ports on the Beryl AXs as 2x1Gbps in a LAG, connected to the switches. Similarly the MT6000 has 2x2.5Gbps and 4x1Gbps ports. 1x2.5Gbps is the uplink to the ISP, but could some combination of the other ports be turned into a LAG?
I already have enough cat5e running in the right places to do this, and the switches all support LAG and LACP, so should be possible to interlink. But is this even possible with the Beryl AX and MT6000 devices? What's the art of the possible here, and is it really worth my effort?
Plan sounds good, What is Glinet AX3000? I dont find one on the internets. If you have 2.5Gbit eth0 DSA system port you can use 2 LAN ports, if it is gigabit you need to use LAN+WAN. Note individual data stream like tcp connection will follow one wire, it will not be totally double speed.
Personally, I don't recommend LAG connections unless you are actually saturating your 1G lines currently. This means that you would need to have either >1G internet speeds and/or a significant amount of high bandwidth local activity that routinely maxes out the available bandwidth on your LAN. And if things aren't configured properly, the LAG can actually introduce problems into your network that can be hard to diagnose and resolve.
Since you have managed switches now, you can monitor the bandwidth through the uplink ports to see if they are indeed saturating.
To be clear, the proposal you have shown is fine and there's nothing wrong with adding LAG. But it won't improve your network performance unless you are indeed saturating one or more of the links.
I'm assuming that as there is a 2.5Gbit port present, the SOC has a 2.5Gbit connection to the switch, and enough performance to drive two bonded 1Gbit connections. And yes, I understand any TCP session stays on a single channel - so I grow capacity but my speed is still capped at 1Gbit for any client.
That's very useful advice. I'd not realised I could monitor the saturation of the links using the switches. School-boy error
Appreciate your comment that the proposal is fine - its useful to know that I have a workable plan, if I do start to experience issues with wireless clients driving the wired network to saturation. But maybe by then it might make better sense to bite the bullet and re-pull my cat5e with cat6a and upgrade the switches again...
That's also an option. Consider fiber at that point (maybe alongside your copper).
But also, if you have the ability to test you cables with a high-ish end cable tester, you might find that you can run faster that 1Gbps over your cat5e cables. Keep in mind that technically, 1Gbps is the rated max, of course, but I actually used a high end Fluke cable verifier and found that I can reach 5Gbps on the 5e runs through my house. I'm only currently running them at 1Gbps, but it's nice to know it's possible to push the limit (although you never know if I will start to experience issues with the higher speeds).
Are these gl.net devices capable of pushing 2 gbit/s with any kind of bond? Aka do these devices have this hardware feature? IIUC as in a worst case all that traffic hits the CPU this can become the bottleneck. I only have experience on x86 and enterprise grade switches... That's the reason I'm asking...
The MT3000 has a 2.5G port and a 1G port. I don't know if you can bond them, so the APs themselves may not benefit unless the switches are upgraded with models that have >1Gbps ports.
However, if the OP is actually saturating the switch uplinks, LAG could help overall network performance to benefit all the other devices connected to those switches. But, per my earlier comment, this is only beneficial if there are clear limitations on the switch uplink ports.
In practice, 802.11ax with 2x2 clients (which are the most common), your practical -effective- throughput tends to be in the 600-750 MBit/s range, so well below the capabilities of wired 1000BASE-T. Yes, we can discuss concurrent access, Mu-MIMO and multiples APs, but for 802.11ax, 1 GBit/s is still a reasonable backhaul network; for 802.11be you might like 2.5GBASE-T (but that's rarely exceeding 1 GBit/s at this point either, be it because of the drivers' infancy (windows 11 24h2 will be the first version to officially support wifi 7) or because many features of wifi7 (MLO and similar) simply aren't available in many contemporary devices, with the current drivers/ firmware, yet).
My cat5e cable runs are mostly under 20m so like you, I may well manage higher speeds, but I'm always wary of running too close to the wind. Bad things happen at the extremes
If I recable then I'd put fibre and cat6a in together. Fibre was prohibitively expensive when I originally ran the cables (turn of the century) but the incremental cost compared to the grief of actually repulling everything makes it a no-brainer ... time moves on!
My old router-turned-into-accesspoint is continuing to intermittently freeze (the hardware is a decade old now) so I'll probably just swap them all out for MT3000s over time for reliability. At that point I can't imagine that I won't at least try to bond one of them to the switch with LAG, just to see if it is possible or not, even if it's not strictly required :slight smile:
Will let you know at that point how I get on...
Thanks for this really useful information. Everything you tell me says that its unlikely I need to worry about the backhaul after all, and as @psherman pointed out, I can monitor the bandwidth utilisation now I have managed switches too, so I'll be able to see if I do start to get near the limits and do something at that point.
Dont Nail me but maybe you are lucky and one or the other lacp or bond mode are supported by the router chip. Your managed switches however should be supporting it. But yes just try it but don't be surprised if the router AP doesn't perform good if all traffic hits the CPU.
