Low 5 GHz performance of Intel Advanced-N 6230

Dear all,

I have been moderately recently hit by a performance drop of Intel Advanced-N 6230 card. I unfortunately do not have the "before" numbers, but I have noticed this since I was able to stream bluray rips over nfs to a wired Raspberry Pi a while ago and now I cannot. This has lead me to begin investigating the issue.
In the end I have used iperf3, testing speeds between RPi and laptop, with laptop running both Windows and Linux. iperf3 on RPi was always the same, on the laptop it was self-compiled iperf3 git master (on Windows I used Cygwin). Please see the observed speeds below:
- Windows: upload to RPi 93 Mbit/s, download from RPi 54 Mbit/s
- Fedora: upload to RPi 26 Mbit/s, download from RPi 48 Mbit/s

This is all on Fedora 20 x86_64. Please find attached the journalctl log for the last boot. Please let me know if more information is needed - I would love to get my network speeds back up to the level where I can watch bluray rips again. Thank you for your support in advance.

Best regards,
Julian

W dniu 24.04.2014 08:10, Julian Sikorski pisze:

Quoted 14 lines

Dear all,

I have been moderately recently hit by a performance drop of Intel Advanced-N 6230 card. I unfortunately do not have the "before" numbers, but I have noticed this since I was able to stream bluray rips over nfs to a wired Raspberry Pi a while ago and now I cannot. This has lead me to begin investigating the issue.
In the end I have used iperf3, testing speeds between RPi and laptop, with laptop running both Windows and Linux. iperf3 on RPi was always the same, on the laptop it was self-compiled iperf3 git master (on Windows I used Cygwin). Please see the observed speeds below:
- Windows: upload to RPi 93 Mbit/s, download from RPi 54 Mbit/s
- Fedora: upload to RPi 26 Mbit/s, download from RPi 48 Mbit/s

Quoted 8 lines

This is all on Fedora 20 x86_64. Please find attached the journalctl log for the last boot. Please let me know if more information is needed - I would love to get my network speeds back up to the level where I can watch bluray rips again. Thank you for your support in advance.

Best regards,
Julian

Hello again,

I have narrowed this down to have regressed between 3.13.6-200.fc20 and 3.13.7-200. Using pointers from Josh Boyer of #fedora-devel, I have managed to restore the previous performance by loading iwlwifi with 11n_disable=8. I am now at 3.14.1-200 with that option enabled and the network is fast as it used to be.
Is such a major performance hit expected from this change? Here is the commit in question:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.13.y&id=45e5cb4f43d33e72ff5f98c80b081eb42e4e4182

Quoted 4 lines

Is such a major performance hit expected from this change? Here is the commit in question:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.13.y&id=45e5cb4f43d33e72ff5f98c80b081eb42e4e4182

Right - I disabled TX APMDU in this patch. This feature gives a big boost in TX performance, but lots of people experienced bugs that disappeared when this feature was disabled. This bug is in the firmware. Since the most common use case is to browse the internet which is more Rx than Tx, I decided to disabled this feature by default and allow people to enabled it using the module parameter.

2014-04-25 9:46 GMT+02:00 Emmanuel Grumbach <[email protected]>:

Quoted 7 lines · Julian Sikorski

Quoted 4 lines

Is such a major performance hit expected from this change? Here is the commit in question:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.13.y&id=45e5cb4f43d33e72ff5f98c80b081eb42e4e4182

Right - I disabled TX APMDU in this patch. This feature gives a big boost in TX performance, but lots of people experienced bugs that disappeared when this feature was disabled. This bug is in the firmware. Since the most common use case is to browse the internet which is more Rx than Tx, I decided to disabled this feature by default and allow people to enabled it using the module parameter.

Any plans to update (fix) the firmware?

Quoted 7 lines

2014-04-25 9:46 GMT+02:00 Emmanuel Grumbach <[email protected]>:

Quoted 6 lines · Julian Sikorski

Quoted 4 lines

Is such a major performance hit expected from this change? Here is the commit in question:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.13.y&id=45e5cb4f43d33e72ff5f98c80b081eb42e4e4182

Quoted 4 lines

Quoted 1 line

Right - I disabled TX APMDU in this patch. This feature gives a big boost in TX performance, but lots of people experienced bugs that disappeared when this feature was disabled. This bug is in the firmware. Since the most common use case is to browse the internet which is more Rx than Tx, I decided to disabled this feature by default and allow people to enabled it using the module parameter.

Any plans to update (fix) the firmware?

I don't work on the firmware level, so I can't really say, but I doubt someone will have the cycles to fix the firmware for these devices. Newer devices (7260 and up) have this feature enabled.

W dniu 25.04.2014 11:51, Emmanuel Grumbach pisze:

Quoted 12 lines · Rafał Miłecki

Quoted 9 lines

2014-04-25 9:46 GMT+02:00 Emmanuel Grumbach <[email protected]>:

Quoted 7 lines · Julian Sikorski

Quoted 4 lines

Is such a major performance hit expected from this change? Here is the commit in question:
https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-3.13.y&id=45e5cb4f43d33e72ff5f98c80b081eb42e4e4182

Right - I disabled TX APMDU in this patch. This feature gives a big boost in TX performance, but lots of people experienced bugs that disappeared when this feature was disabled. This bug is in the firmware. Since the most common use case is to browse the internet which is more Rx than Tx, I decided to disabled this feature by default and allow people to enabled it using the module parameter.

Quoted 5 lines

Quoted 2 lines

Any plans to update (fix) the firmware?

I don't work on the firmware level, so I can't really say, but I doubt someone will have the cycles to fix the firmware for these devices. Newer devices (7260 and up) have this feature enabled.

Can this feature be disabled on a per-device basis? It clearly works for 6230 and is a huge performance hit when gone.