Re: [Madwifi-devel] sample rate control buggy

[Benoit P. <ben...@fr...>]

Indian Mogul wrote:

> Ans: The algo misjudges the underlying wireless medium. This 
> is dictated by the history of event in the ewma technique.
> If rate (i) = rate_now * x + (1-x) rate_history
> try making x = 0.8 (high number) so that it puts more value to 
> what it see now.. However, there is a big side effect in doing so.. 
> you might be opening yourself to transient variations of wireless
> channel. So, think before you rush to change 'x'... The actual solution 
> is more detailed and involves using a better channel model. 

I will try your suggestions. However, from what you said and from what I 
learn reading about ewma is that ewma on all rates should be done using 
the same sample period. Otherwise, we end up comparing average 
transmission time from different sample period, which is currently what 
occurs in the code. For instance, the current rate has a lot more sample 
that any other rates. Am I right?

If so, maybe we should fallback to the original definition as computing 
the average for the last 10s.

What is a "better channel model" ?

Best regards,
Benoit