My take on using Link Aggregation Group and OpenWrt

Community Builds, Projects & Packages
1

Hello,

Since I wanted to use LAGs on my home OpenWrt setup (router + access point) but found little documentation, I managed to do it and I wrote a post on my blog with all the necessary steps. (Ad-free blog, I get no monetary return).

I configured LAG on my nanoPi R5S router connected using SFP+ to the switch and to the Netgear WAX206 access point. It’s quite easy but I really found little documentation, especially for the new DSA architecture, it’s a shame because the improvement is not super but there is!

Here’s the post:

Homelab setup v6.5: New router, LAGs, SFP+ and some optimizations

Hope this helps, and don't break the forum rules!

2 Likes
2

I had no idea what's a LAG and I had to open the link in the hope that it explains the term (which it does). I suspect that many/most people wouldn't know what this is either and wouldn't bother to click on the link. So I would suggest to replace the abbreviation in the title with "Link Aggregation Group".

1 Like
3

I once tested LAG on the same router. Even LAG offload worked for me, with minor issues. I think I need to bring back this topic. Here is my last version.

4

Done, thank for the input =] hope my poste explained the LAG!

1 Like
5

Neither the NanoPi R5S nor my MikroTik switch has the MT7530 chip, I don't understand well what you mean, sorry.

6

Netgear WAX206 has an MT7531 switch.

7

Oh correct, sorry I forgot it :sweat_smile: But the LAG is working on my Netgear WAX206, so I still don’t understand well what’s the point of the PR on GitHub, only the hardware offload support to the LAG on MT7530/1? Because LAG it’s working very very well on my WAX206, and the load on the CPU is almost zero. I haven’t investigated but look like it’s already working. Sorry for the question but I’m not sure I understood correctly.

8

With hardware offloading, traffic can pass through the switch without CPU intervention. This applies to LAG with bandwidth aggregation. If you have failover, it doesn't really change much.

9

Consider a scenario where one host is connected to WAX206 with two 1G cables. The second host is connected to a 2.5G port. Then, by running iperf with multiple connections between these two hosts, you can achieve 2Gbps full duplex. Without offloading, the CPU processes all traffic and saturates at around 2Gbps in simplex.

10

Ok, don’t have the hardware to test this, but now I can already saturate 2gbps, if I run iPerf from another machine with LAG (as written in my post) the test saturates both the ports = 2gbps.

This is why don’t really understand the hardware offloade issue.

11

Run it in duplex(bi-directional) and you still will be capped at 2 Gbps while link saturation is 4Gbps(2Gbps duplex)

1 Like
12

Ah okay because you mean using full duplex/bidirectional testing, now I understand (I think the nanoPi R5S has the same issue).

Anyway using 1gbps up and 1gbps down simultaneously is enough, my main goal is to distribute the streams, rather then having a very high throughput.

13

You don’t need LAG for 1Gbps duplex btw

14

Yes correct, this is why I wrote “my main goal is to distribute the streams, rather then having a very high throughput”

Anyway the hardware offload for the Netgear wax206 switch would be nice generally speaking.

1 Like