Would you mind trying without firewall active? Getting 332mbps with a 1404mbps link speed is quite bad. I would have expected at least half of link speed. The netfilter stack likely is sucking a lot of CPU cycles causing a big drop in performance.
If true, we really need some form of h/w offload implemented for the ath10k for the R7800.
The client is relatively easy to test with as it's a non production build server, but the router itself is production and therefore difficult to test with particular without a firewall.
Thanks. Seems we still have a huge bottleneck somewhere.. For reference an ipq40xx (2:2 device) links at 866 and gets iperf3 runs in 500mbps+ range (only when cpufreq set to max though).
Did you notice cpu utilization through those runs? And did you test with parallel streams (with -P flag). Don't bother with retesting if you don't have the data, I was just wondering. Thanks again
Anyone else concerned about the move to -ct? Unfortunately for me it does not work well with my clients. The problem I have is with 2 IoT devices (one Nest E thermostat and one based on MediaTek LinkIt Smart 7688). It seems that when they go into power save mode they lose connection, they don't respond to arp requests and even when they are awake there's packet loss during ping. I've also seen a Macbook Pro 2018 have problems a few times, it's connected but the traffic just stops, I have to disconnect and reconnect to wifi. None of these problems with the "old" builds. Though on the old builds the router sometimes just randomly freezes or reboots. I don't think it's a hardware problem as it does the same on 2 units. They don't run hot, they just do AP mode, all other services disabled, traffic is fairly light.
CT seem more stable now, I am using it myself. However I havent had one issue at all with the official firmware. So I don really see the benefit using CT when it seems to be falling behind vs the official firmware too.
Sure, CT might be faster, have more "features", but what does that help if the firmware crashes for some people.
Its not like someone is forcing you to use CT, simply switch to non CT firmware.
The main reason behind the switch is that Ben will actually fix the issues if reported and you know what has changed between versions, with QCA you have no idea what has changed
Well, I don't think most people know where to get the official firmware and/or how to install it.
Installing openwrt is quite easy, changing the firmware is also easy. The problem is that I don't think they know it's a separate firmware in the first place.
Well, the thing is that like non CT version, it will work for 99% of people.
Those that have issues tend to google or ask here on forum.
Either way there are gonna be some issues, but CT versions are much better from a maintance stance
I have had long standing issues with my R7800 crashing randomly when using 5GHz wifi, once I connect with my laptop and stress the connection it would just die withing a few seconds and reboot.
I've now tried this build: https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=319755 (KONG 19 2019-Q2 / LuCI openwrt-19.07 branch (git-19.190.55584-ab90938)) and with that suddenly my 5GHz wifi works perfectly!
I'm wondering what might be different, there are a few patches in https://www.desipro.de/openwrt/sources/ but I'm not sure how they are stopping the wifi from crashing.
When you say "when using 5GHz wifi" does that mean that everything works well on 2.4GHz? Have you been able to catch any messages in the log that might indicate why the router crashes?
Yes, everything works perfectly on 2.4GHz, I've been using the router for years that way. But I could crash it reliably when using 5GHz from my laptop, just run a speedtest and tada it reboots.
I've never managed to get any useful logs, I've even tried reading dmesg in a loop over a wired connection but it always dies before logging anything. And since it reboots I'm not getting any logs afterwards.
Hmm, strange. I assume that you have tried both ath10k and ath10k-ct?
I think that build is using the regular ath10k drivers and I suspect they are the latest (OpenWRT is using the older ones from linux-firmware).
Yes, tried both a lot, I was pretty certain by now that this was some kind of hardware issue.
I have zero problems with wifi clients on R7800 in 5 GHz. Marvell, Intel, Realtek, other Qualcomm, all just work well.
There is now a apparently working solution for the crashes with the old mainline ath10k:
I built and tested with r10825-2ceee0e023.
The router stays alive and wifi works with the old mainline ath10k, again.
(reference to FS#2414 )
Confirmed, r10825-2ceee0e023 && nbg6817 && kernel v4.19.67.
As you noticed "SWBA overrun on vdev 0, skipped old beacon" is going rampant, but that's a different issue.
I tested also with firmware-5.bin_10.4-3.10-00047 from kvalo's repo (as it is slightly newer than the 10.4-3.9.0.2-00046 in linux-firmware), but that does not help to the SWBA overrun bug.
Is it normal for the buildbot 19.07 to have pending build requests staying in the queue for days?
Depends.
The build request count can be faulty in any case, as the packages buildbot (to which you are apparently refferring to) has changes from many feeds, so the listing is not quite right always.
If you look at the log, you can notice that packages for R7800 have been built twice yesterday, but now the latest attempt has failed.