192.168.90.4 is the R7800
192.168.90.5 is a Linksys WRT 32x running OpenWrt 22.03
The same test with a Linksys EA8300 on 23.05 and WRT 32x on 22.03 and with R7800 on 22.03 and WRT 32x on 22.03 give near gigabit throughput on both directions.
Yes there is a very expensive TP-Link managed 24 port gigabit + fibre switch in the middle and I am using exactly the same ports for the tests. With the R7800 on 22.03 I get full speed, upgrade the R7800 to 23.05 and I get a quarter speed on the second test
The switch does not have any bearing on the results.
Here are the test results from the same two routers using exactly the same cables and switch ports, only this time the R7800 was downgraded to OpenWRT 22.03
Did you check if the CPU is running at 1725 Mhz?
I think nowadays it should use the performance governor.
Is irqbalance used?
I have seen other threads from users complaining about less performance on newer builds, which was attributed to newer kernel, DSA, nftables etc. but I have not seen figures as low as yours.
My R7800 does slightly over 500 Mb/s without any off loading or other tweaking it is still using a snapshot from before we had 23.05 but that should be comparable (In the past I could get much better throughput but I am talking about Kernel 4.9, iptables, swconfig etc).
irqbalance is installed, does it require any special configuration? And where would it be configured?
I noticed quite a downward trend in OpenWRT with newer releases. They are trying to keep up with new kernel releases which breaks most devices. I would de-prioritize new kernels in favour of performance and compatibility with well established devices. The linux kernel is conceived for servers and I think that kernel developers don't pay enough attention to SOCs and legacy hardware compatibility, they are probably a nuisance to them. If I were working on a project like OpenWRT I would pick an optimal kernel release, work on it, optimize it and stick to it, applying only security and bug fixes. Until the next major OpenWRT release. All major Linux distributions are on some older kernel that gets constantly patched, but never uplifted to new releases. OpenWRT is the exception. It is as if OpenWRT developers were obsessed with running after the latest kernel release.
Besides, a router should provide max performance for network devices as a priority out of the box. If network performance requires any tweaking from a default configuration then something is really wrong.
Yes, I have seen the reports. And R7800 is not the only fast and popular router degraded by newer OpenWRT releases. Linksys WRT mvebu are another family of excellent routers that work best with OpenWRT 19. Lantiq based BT Home Hub 5 is another one that works best on OpenWRT 19. BT Home Hub 5 is still one of the best in WiFi range despite its age. It is the only one that can handle all my remote WiFi cameras on WiFi 2.5g around the house and garden. Newer routers with space age looks and NASA style antennas can't get stable connections.
The lack of regression testing and leaving behind popular routers are symptoms of a degrading strategic direction for OpenWRT. Please see my comment above regarding the continuous Linux kernel upgrades for an OpenWRT release. Maintenance for OpenWRT 19 should be resumed until the newer releases are properly fixed.
If you’re unhappy with OpenWrt: fork the code and demonstrate that your fork does better.
I see that people in the community like to always run the latest and greatest hardware on the newest kernel and user space software. The strategy to get to most recent LTS kernel looks good to me. Linux 6.1 is the current LTS kernel. It’s used in master branch. Linux 5.15 LTS will be the choice in OpenWrt 23.05.
If you are unhappy with this: please fork OpenWrt 19.07 with Kernel 4.14. This kernel will go out of support in January 2024. The community will be happy seeing you providing security fixes for both kernel 4.14 and your OpenWrt 19.07 fork by yourself, for free. Please do it.
Exactly, it's very easy to comment and demand stuff done.
I've tried a few years ago to fork 17.01 for some different reasons - support for 4mb flash devices and it was a pain just to start doing it. I've even created some test builds and the first reports disapointed - people just said it doesn't work - more "reasons" to carry on...
Indeed I am unhappy with the direction OpenWRT is taking, otherwise I would not be giving this feedback, but your response hinders constructive dialogue.
If you want to listen and consider mine and other people's concerns, I am happy to contribute with my advice.