Ath10k and mesh (getting just 1/3 of expected speed on Snapshot as of 16.apr. 2023)

I’m interested in other peoples experiences regarding 802.11s with/without batman.

On of the problems I’m facing is that I only get one third of the speed between a wireless 80mhz 802.11s/batman connection between 2 r7800 (qca9884) when using snapshot compared to 22.03.4. Seems almost they’re using 1mimo instead 3mimo. 22.03.4 give me about upwards 50MBs while Snapshot 16-16.5MBs.

My Lyra (qca9886) running snapshot seems to give the right speeds (about 2mimo, 30-33MBs incl batman overhead) note that this is using lyra as a node with a R7800 as the main router. I haven’t yet tried Lyra to Lyra.

(OT but this is nowhere the speeds I get from my iphone @ 2MIMO which is saturating a 500/500 line connected to the main R7800. This is ofc without the Batman overhead and all but daym :slightly_smiling_face:)

Could there be a regression somewhere?

1 Like

Are you using batman iv or batman v? I noticed a significant improvement in my setup after switching to Batman v.

I do run a setup with a lyra as my main router with a 500/20 link and my lyra gets about 325 down with irqbalance, software offloading and packet steering. I haven’t tried changing from ondemand to performance for the governor. After learning more about governors, I’m pretty sure the lyra can utilize it as an ipq40xx, but I’m not positive.

I’ve got two Lyra’s, a C2600, and two RE350K’s in my current setup and I get around 50mbps download on all my nodes, except the main lyra and the c2600 which has an Ethernet backhaul and I get much better speeds. The rest of the setup uses a dedicated 5ghz radio and although everything looks great as far as speed between devices, as mentioned, they top out at 50. I use snapshots for my Lyra’s and my C2600, while the RE350K’s use 22.03 and only offer a 2.4 radio to me cause I dedicate their 5ghz radio to the backhaul.

My PC is wired to one r7800 and my NAS is wired to the other r7800: I just switched from iv to v and the speeds are the same. Maybe it’s less fluctuating, but I’d have to retest that a few times to be certain.

I have no idea how the Lyra measure as a main router cos I didnt test that. Seeing that my R7800 managed over 500mbs connected to my iphone (@2mimo 866mbs) I would expect the same from the Lyra but maybe it doesn’t and iphone and R7800 is just a special thing lol. Anyways, I think you’d be better off with your c2600 as a main router, coz it being the main router you’ll be fine with just using 1 5ghz radio.

I also have a TP-Link RE-650 (mt7621/mt7615e) and that gave me about 225-230mbs as a node measured with speedtest on my phone, which means about 450-460 in total using just 1 radio, so just a tad slower Lyra which can use a dedicated 5ghz radio for mesh/ap. (I do find the main radio on the Lyra has better reach @ 23dB compared to 30dB using the other radio… which is still really bad.. I thought this was fixed.. This was with ct firmware, maybe it’s different with the normal firmware)

Also the ct firmware (for all my ATH10K devices) seems to work fine with MT7621 devices, but not with other ATH10K devices. There’s so many variables at this point, so it’s hard to pinpoint what and whatnot, but at least with the CT firmware it’ll be anything from almost stable (until it’s not) to very unstable depending on time of day, OpenWrt version or some other random cause. So in the end I gave up on that and just use the normal firmware with the ct drivers. (The wave 1 ct firmware doesn’t support mesh at all while the wave 2 firmwares kinda do, but I’ve come to the conclusion «not really»)

A lot of blabbering but the main point here is still if others can chime in on their experiences using snapshot images vs stable images on ath10k devices running mesh.

With only stationary devices involved, WDS/ 4addr will do just as well, typically at better performance (and easier to set up).

Not badly, but worse than a r7800 (slower SOC and slower wireless - but their tri-radio design makes them perfect repeaters or mesh participants).

Oh that was just to clarify how I measured the speed in this particular case (meaning the mesh performance itself). I do use wireless devices as well. Actually my main PC is using it's own SSID on the R7800 connected to my NAS (Windows reports peaks of over 70MBs/560mbs@2mimo 866mbs), while my HTPC is connected by wire to the main R7800. But this is besides the point. I know I can get better speeds by using, say relayd, something close to 70MBs/560mbs instead of 47-50ish MBs/400mbs. (just by using 80mhz bw). There's a lot of overhead in using a mesh, but anything close to 50MBs will work out just fine until we're all on tri-radio wifi7 160mhz bw on 6ghz without the DFS bs.

Apart from this, for me it's just as easy to set up a mesh, if not easier than using relayd or WDS. The beauty of a mesh is the L2 bridging, that I can always reach each node, the self healing nature. (I can just fire up a new node whenever and it'll be part of the network in no time) plus I can do batman-vlan.

You'd think that, and that's why I bought them in the first place. But I found the second radio still kinda bad, so I haven't yet done any serious benchmarking on them.

A little OT but I whipped out my Lyra running this weeks snapshop (coz it's using DSA and 22.03.4 does not) and did a non scientific test.
330mbits with peaks @ 360mbits using the upper channels radio, as an AP. Packet steering on. Irqbalance had no effect whatsoever. the 360mbit peak seems to be limited by the fourth cpu core that hits 100%, while overall load is just 130ish % out of 400. Peaks of 220mbits with routing, never saturated any core, so that's a bit weird.

The lower channel radio gave maybe 20mbits more when routing. While the speeds as an AP were a little lower (peaks at about 300mbits, never fully utilising the fourth core) when using the lower channel radio, possibly because of more interference. (Although they were Idle, I didn't turn off my other devices and mesh that use the same channels)

Anyways, it doesn't seem anyone else is using R7800 in mesh and can chime in their experiences. I'll just hope this gets fixed in time for the next major OpenWrt version.