Odroid M1, Network Adapter Performance, in jammy

[rvalle]

Hi!

I am looking into OdroidM1 network performance with kernel 6.6.47-current-rockchip64 ( 24.8.1 )

For some reason network performance is x10 faster (upstream) in bookworm and noble, when compared to jammy.
Downstream performance is fine, delivering the same performance in jammy and the other distributions.

Since all of them are using the same kernel, I assume that there must be some configuration that jammy is missing, or?

Any idea about what could be going on?

[rvalle]

 

@Igor , thanks for your comment:

 

Quote

Jammy, Noble, Bookworm ... are not different in this. Pay attention to kernel version. Its a known regression, also found on official Hardkernel images. No solution yet.

 

I am getting some variation, but perhaps all of them are degraded anyway.

 

While I wait for the upstream fix, is it possible to build armbian on the last working kernel version?

 

How would that go?

 

[rvalle]

@Igor this recent fix seems like it: https://lore.kernel.org/all/E1sbJvd-001rGD-E3@rmk-PC.armlinux.org.uk/#r

 

apparently the driver was decoding the duplex mode incorrectly.

[usual user]

1 hour ago, rvalle said:

this recent fix seems like it

Allready landed in mainline, you get the fix out-of-the-box since v6.11-rc3.

[Igor]

4 hours ago, rvalle said:

this recent fix seems like it

 

If patch was not back-ported already by kernel.org to stable branches, pull them into CURRENT and EDGE with a patch and we can ship an update right away:

 

https://github.com/armbian/build/tree/main/patch/kernel/archive/rockchip64-6.10

https://github.com/armbian/build/tree/main/patch/kernel/archive/rockchip64-6.6

[Igor]

Before:
 

[  5]   0.00-1.00   sec  63.5 MBytes   532 Mbits/sec  250   22.6 KBytes       
[  5]   1.00-2.00   sec  71.2 MBytes   598 Mbits/sec  234   21.2 KBytes       
[  5]   2.00-3.00   sec  68.6 MBytes   576 Mbits/sec  246   25.5 KBytes       
[  5]   3.00-4.00   sec  68.5 MBytes   575 Mbits/sec  250   26.9 KBytes       
[  5]   4.00-5.00   sec  68.6 MBytes   576 Mbits/sec  243   18.4 KBytes       
[  5]   5.00-6.00   sec  67.9 MBytes   570 Mbits/sec  252   29.7 KBytes       
[  5]   6.00-7.00   sec  68.9 MBytes   577 Mbits/sec  248   28.3 KBytes       
[  5]   7.00-8.00   sec  67.2 MBytes   564 Mbits/sec  250   28.3 KBytes       
[  5]   8.00-9.00   sec  70.1 MBytes   589 Mbits/sec  242   15.6 KBytes       
[  5]   9.00-10.01  sec  69.8 MBytes   580 Mbits/sec  243   19.8 KBytes  

 

After:

 

[  5]   0.00-1.00   sec  87.4 MBytes   732 Mbits/sec  143   24.0 KBytes       
[  5]   1.00-2.00   sec  92.1 MBytes   773 Mbits/sec  153   65.0 KBytes       
[  5]   2.00-3.00   sec  89.1 MBytes   748 Mbits/sec  165   39.6 KBytes       
[  5]   3.00-4.00   sec  89.2 MBytes   749 Mbits/sec  167   31.1 KBytes       
[  5]   4.00-5.00   sec  85.8 MBytes   719 Mbits/sec  180   38.2 KBytes       
[  5]   5.00-6.00   sec  88.6 MBytes   743 Mbits/sec  173   12.7 KBytes       
[  5]   6.00-7.00   sec  88.4 MBytes   741 Mbits/sec  171   59.4 KBytes       
[  5]   7.00-8.00   sec  89.9 MBytes   754 Mbits/sec  158   28.3 KBytes       
[  5]   8.00-9.00   sec  89.0 MBytes   747 Mbits/sec  164   21.2 KBytes       
[  5]   9.00-10.01  sec  87.6 MBytes   729 Mbits/sec  182   39.6 KByte

 

[Igor]

Made a PR https://github.com/armbian/build/pull/7245

[rvalle]

 
If patch was not back-ported already by kernel.org to stable branches, pull them into CURRENT and EDGE with a patch and we can ship an update right away:
 
https://github.com/armbian/build/tree/main/patch/kernel/archive/rockchip64-6.10
https://github.com/armbian/build/tree/main/patch/kernel/archive/rockchip64-6.6

Sorry, have never done this... Will check how you did.

Sent from my 2201116SG using Tapatalk

[rvalle]

Made a PR https://github.com/armbian/build/pull/7245

Nice!

Sent from my 2201116SG using Tapatalk

[rvalle]

On 9/14/2024 at 6:45 PM, Igor said:

pull them into CURRENT and EDGE with a patch and we can ship an update right away:

 

@Igor I am trying to understand the publishing method. How are patches turned into updates?

I guess updates are pushed from armbian-firmware repo, or? But I cant see the relationship between the builder and the firmware repos....

You mentions updates go out right away. how does it work?

[Werner]

Updates are published via our own apt repository. Those packages are generated with the same build framework you can find on Github plus some additional scripts which are there as well (probably in a different repo though). Maintainers can trigger an update to publish a new kernel version so end-users can update with a simple apt upgrade.

[Igor]

1. Script build all changed packages every 4 hours or manually in-between:

https://github.com/armbian/os/actions/workflows/complete-artifact-matrix-all.yml

2. If this is successful repository rebuild follows (rolling and stable)

https://github.com/armbian/os/actions/workflows/repository-update.yml

3. Once finished, packages are pushed to rolling release repository http://beta.armbian.com (which is stable enough for this hardware and most of end users). As kernel covers many devices, pushing to default stable repo https://apt.armbian.com happens manually after observing this https://github.com/armbian/os?tab=readme-ov-file#latest-smoke-tests-results

 

None of other distributions or distributions provided by vendors have this kind of quality control. Sadly, not even this is enough, but better we can't afford.

[rvalle]

On 9/17/2024 at 11:30 AM, Igor said:

None of other distributions or distributions provided by vendors have this kind of quality control. Sadly, not even this is enough, but better we can't afford.

 

I agree, yes. It is very nice and Agile.

 

I can see the new kernel release .51 in beta apt repo.

 

Will test it in the environment I saw the problems.

 

R.