net80211 ratectl proof of concept

Rui Paulo

unread,

Mar 31, 2010, 9:05:52 AM3/31/10

to Sam Leffler, Andrew Thompson, Weongyo Jeong, Bernhard Schmidt, freebsd-net

Hi,
I've started developing a ratectl framework for net80211, loosely based on what DragonFly has. Right now only one driver has been ported, but I would like your feedback before continuing.

The objective is to, eventually, have all the ratectl stuff (amrr, sample, onoe(?) and rssadapt) in net80211 so all drivers can use it. We can also select which ratectl modules are built in the kernel config file.
The framework support changing the current ratectl is out of scope for this patch.

You can find the patch here:
* http://people.freebsd.org/~rpaulo/ratectl.diff

Only the ral driver and the AMRR rate control algorithms were ported.

Some comments:
o The rate control calls now dereferences several pointers and some inline functions are now real functions. I wonder how much this impacts performance and what we can do to solve it.

o I wished there was a better way to do the IEEE80211_AMRR_SUCCESS / IEEE80211_AMRR_FAILURe call.

o Some other stuff can also be `const'

o I create ieee80211_ratect.[ch] to avoid polluting other files

o I moved the AMRR parameters inside amrr_init() on purpose. The drivers we have now only specify a different interval and I plan to add export amrr_set_interval() via the ratectl framework later.

I would like very much to see this in, unless there's a strong impending argument.

Thanks,
--
Rui Paulo

Bernhard Schmidt

unread,

Mar 31, 2010, 10:04:15 AM3/31/10

to Rui Paulo, freebsd-net, Antoine Brodin

On Wed, Mar 31, 2010 at 02:31:30PM +0100, Rui Paulo wrote:
> On 31 Mar 2010, at 14:24, Antoine Brodin wrote:
> > This looks great!
> > Is there specific reasons to use pointers and not ints for some
> > arguments of foo_tx_complete() and foo_tx_update()?
>
> Not really. I'll probably switch them to ints at some point.

Hmm, I somehow like the idea of not passing integers here. Looking at
ath_rate/sample it passes a struct. I don't know whether this can be
rewritten though.

--
Bernhard

Rui Paulo

unread,

Mar 31, 2010, 9:31:30 AM3/31/10

to Antoine Brodin, freebsd-net

On 31 Mar 2010, at 14:24, Antoine Brodin wrote:

> Hello,

>
> This looks great!
> Is there specific reasons to use pointers and not ints for some
> arguments of foo_tx_complete() and foo_tx_update()?

Not really. I'll probably switch them to ints at some point.

--
Rui Paulo

Rui Paulo

unread,

Mar 31, 2010, 10:05:27 AM3/31/10

to Bernhard Schmidt, freebsd-net, Antoine Brodin

It can be rewritten to pass ints, from what I understand.

--
Rui Paulo

Bernhard Schmidt

unread,

Mar 31, 2010, 1:06:48 PM3/31/10

to Rui Paulo, freebsd-net, Antoine Brodin

Ok great, looks fine to me.

Do you intend to commit this in one huge chunk (quite a few drivers
affected) or create a branch for that?

--
Bernhard

Rui Paulo

unread,

Mar 31, 2010, 1:14:21 PM3/31/10

to Bernhard Schmidt, freebsd-net, Antoine Brodin

I wanted people to comment on the general framework.

I wasn't planning on creating a branch, because I didn't feel it was necessary. Porting AMRR code to this framework isn't hard and I can deal with any fallout.

--
Rui Paulo

Rui Paulo

unread,

Apr 5, 2010, 12:54:20 PM4/5/10

to Rui Paulo, freebsd-net, Weongyo Jeong, Andrew Thompson, Bernhard Schmidt

I've ported all the drivers but I can't test them all. You can read the patch at the same URL.

I would like to commit this soon, though.

Regards,
--
Rui Paulo

Weongyo Jeong

unread,

Apr 6, 2010, 9:19:25 PM4/6/10

to Rui Paulo, freebsd-net, Andrew Thompson, Bernhard Schmidt

It looks it's what I really want to see. Please go forward.

regards,
Weongyo Jeong