[question] Rate control algorithm for mesh

[question] Rate control algorithm for mesh

[ferran]

Hello everybody,

I would like to know if there is a transmit rate control algorithm out
there that takes into account the destination MAC address. My setting is
like this:

[node0]·  ·  ·  ·  ·[node1]············[node2]

Where the spaced pattern means a poor connection and the densely dotted
pattern means a good connection.

The problem here is that node1 oscillates between fast rates and slow
rates, as when transmitting to node0 it must reduce the rate, and when
transmitting to node2 it increases the rate since it must take profit of
the channel. I am therefore getting many re-tx between n1 and n0, and
n1-->n2 isn't as fast as it could.

Currently I am using the PID-based RC, but I'm afraid that this problem
isn't adressed at Minstrel neither.
Am I wrong?

Thank you for your excellent work!
Ferran

Re: [question] Rate control algorithm for mesh

[Felix Fietkau]

On 2015-11-16 17:02, ferran wrote:
> Hello everybody,
> 
> I would like to know if there is a transmit rate control algorithm out
> there that takes into account the destination MAC address. My setting is
> like this:
> 
> [node0]·  ·  ·  ·  ·[node1]············[node2]
> 
> Where the spaced pattern means a poor connection and the densely dotted
> pattern means a good connection.
> 
> The problem here is that node1 oscillates between fast rates and slow
> rates, as when transmitting to node0 it must reduce the rate, and when
> transmitting to node2 it increases the rate since it must take profit of
> the channel. I am therefore getting many re-tx between n1 and n0, and
> n1-->n2 isn't as fast as it could.
> 
> Currently I am using the PID-based RC, but I'm afraid that this problem
> isn't adressed at Minstrel neither.
> Am I wrong?
Rate control instances are always per sta (so per 802.11 destination).
The problem you are describing either doesn't exist, or is something
different from what you're describing.
I would recommend running a test with latest mac80211 and minstrel(_ht),
it should perform much better than PID.

- Felix

Re: [question] Rate control algorithm for mesh

[ferran]

On 16/11/15 19:16, Felix Fietkau wrote:
>
> Rate control instances are always per sta (so per 802.11 destination).
I totally ignored that.
> The problem you are describing either doesn't exist, or is something
> different from what you're describing.
> I would recommend running a test with latest mac80211 and minstrel(_ht),
> it should perform much better than PID.
>
> - Felix

In this case, I'll repeat the experiments with minstrel enabled. We have
already planned it.

Thank you a lot.
Ferran