A few questions about iwlwifi and its rate control mechanism

A few questions about iwlwifi and its rate control mechanism

[Tambet Arak]

Hi all,

I'm not really a Linux developer, more of a curious user, but the
wireless developers mailing list seems appropriate for my questions.
Please bear with me if they don't make much sense, as I'm a complete
novice when it comes to the wireless subsystem.

In any case, I was looking into how to optimally tune wireless
reception at work, since the APs are aging and only support up to MCS7
(65 Mb/s), so any improvements would help. I'm using a HP laptop with
a Intel Centrino Advanced-N 6235, driven by iwlwifi / iwldvm (vanilla
Linux 4.14.8). I noticed that `iw dev wlo1 station dump` generally
reports a 58.5-65 Mb/s rx bitrate, but when the connection is taxed
(e.g. downloading a file), it drops.

Since the only other OS on my machine is Windows, I tested by
downloading a large file there and generally the average throughput
was slightly higher. I'm not a Windows user, so I don't have a clue if
`netsh show interfaces` reports a correct effective bit rate, but it
was always fixed to 65 Mb/s in my tests.

This led me to digging around in iwlwifi. It seems that iwlwifi
registers its own rate control algorithm (iwl-agn-rs) with the
mac80211 subsystem and this is hardcoded.

Hence a few questions:

- Why does iwlwifi have / need its own rc algorithm, instead of using minstrel?
- It seems that iwl-agn-rs is not configurable by the end user. Would
there be any interest in development for iwl-agn-rs configurability,
e.g. being less / more conservative about lowering rates? I'm mostly
interested in rx performance - I'm making the assumption here that rx
rates are negotiated with the AP (as opposed to being forced).
- Is there any documentation regarding iwlwifi or its rc algorithm
implementation? If so, I would kindly appreciate pointers.

Cheers,

Tambet

Re: A few questions about iwlwifi and its rate control mechanism

[Emmanuel Grumbach]

Hi,

On Thu, Jan 18, 2018 at 11:09 AM, Tambet Arak <tambet.arak@gmail.com> wrote:
> Hi all,
>

[snip]

>
> In any case, I was looking into how to optimally tune wireless
> reception at work, since the APs are aging and only support up to MCS7
> (65 Mb/s), so any improvements would help. I'm using a HP laptop with
> a Intel Centrino Advanced-N 6235, driven by iwlwifi / iwldvm (vanilla
> Linux 4.14.8). I noticed that `iw dev wlo1 station dump` generally
> reports a 58.5-65 Mb/s rx bitrate, but when the connection is taxed
> (e.g. downloading a file), it drops.

What do you mean by 'drop'. Disassociated? Or associated but no traffic?
Association state can be checked with iw <iface> link

>
> Since the only other OS on my machine is Windows, I tested by
> downloading a large file there and generally the average throughput
> was slightly higher. I'm not a Windows user, so I don't have a clue if
> `netsh show interfaces` reports a correct effective bit rate, but it
> was always fixed to 65 Mb/s in my tests.
>
> This led me to digging around in iwlwifi. It seems that iwlwifi
> registers its own rate control algorithm (iwl-agn-rs) with the
> mac80211 subsystem and this is hardcoded.
>
> Hence a few questions:
>
> - Why does iwlwifi have / need its own rc algorithm, instead of using minstrel?

Because Intel devices' firmware works differently from atheros for
which minstrel was designed.
This is one of the main reasons.

> - It seems that iwl-agn-rs is not configurable by the end user. Would
> there be any interest in development for iwl-agn-rs configurability,
> e.g. being less / more conservative about lowering rates? I'm mostly
> interested in rx performance - I'm making the assumption here that rx
> rates are negotiated with the AP (as opposed to being forced).

Not really. iwl-agn-rs is very old and serves EOL'ed devices. Newer
devices have a algorithm that is not very different, but it was more
tested.
It shouldn't be too hard to forbid high rates though. You can just put
some limits in the existing code if you want.

> - Is there any documentation regarding iwlwifi or its rc algorithm
> implementation? If so, I would kindly appreciate pointers.

Well... System requirements are typically confidential, but you can
always dig in the code.

Re: A few questions about iwlwifi and its rate control mechanism

[Tambet Arak]

I see. I might try playing around with hacking iwl-agn-rs, in that case.

> What do you mean by 'drop'. Disassociated? Or associated but no traffic?
> Association state can be checked with iw <iface> link

Sorry, that was ambiguous. I meant that the rate is decreased to e.g. MCS3-5.

Tambet

Re: A few questions about iwlwifi and its rate control mechanism

[Emmanuel Grumbach]

On Thu, Jan 18, 2018 at 12:54 PM, Tambet Arak <tambet.arak@gmail.com> wrote:
> I see. I might try playing around with hacking iwl-agn-rs, in that case.
>
>> What do you mean by 'drop'. Disassociated? Or associated but no traffic?
>> Association state can be checked with iw <iface> link
>
> Sorry, that was ambiguous. I meant that the rate is decreased to e.g. MCS3-5.
>

Oh - that can happen because of lower quality radio conditions.