What I said was when MRR is not enabled, b43 will always use 1 Mbps as
fallback rate, which will significantly reduce the throughput for some
cases.  For a channel with good quality, 54 Mbps may be the best
throughput bit rate.  After two retries fail with 54 Mbps, the bit
rate will drop to 1 Mbps.  But in fact, 48 or 36 Mbps may work well as
fallback rate...

Anyway, I just want to confirm it is what you are doing in the b43
driver and minstrel and I got the correct answers.  Thank you!

-Bo

On Wed, Apr 29, 2009 at 1:19 PM, Michael Buesch <[email protected]> wrote:
> On Wednesday 29 April 2009 22:13:54 Felix Fietkau wrote:
>> Bo Han wrote:
>> > Hi Felix,
>> >
>> > I just want to confirm that when b43 driver set max_rates of
>> > ieee80211_hw to be 2, minstrel will think that MRR is not enabled and
>> > thus will only provide the best throughput bit rate.  There is NO
>> > fallback bit rate calculate in this special case.
>> >
>> > Also, is it possible to modify the code to support just
>> > one-fallback-rate-only case for drivers like b43?  I think minstrel
>> > should be flexible and provide the fallback rates based on the value
>> > of max_rates. Currently, there will be either no fallback rates or 3
>> > fallback rates, right?
>> >
>> > Feel free to correct me if I am wrong.
>> Minstrel will only start properly using fallback rates if it has
>> 4 rates. Using 3 rates would be workable too, but there's not that
>> much that can be done with 2 rates. The reason for that is that
>> it makes sense to have one rate always use the lowest rate, to
>> make sure that when the radio environment changes, it does not
>> produce high packet loss during the time that minstrel takes to
>> readjust the rate statistics.
>>
>> If the driver does not announce 4 rates, minstrel defaults to
>> always specifying the lowest rate as fallback rate, regardless
>> of whether a fallback rate was actually announced by the driver or
>> not.
>
> Ok that makes sense. I understood Bo Han in a way that minstrel would not
> define a fallback at all, so the variable was dangling. That wouldn't be
> acceptable, of course. 1M is OK, of course.
>
> --
> Greetings, Michael.
>
_______________________________________________
Bcm43xx-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev

Reply via email to