Re: [Madwifi-devel] AMRR rate control algorithm causes stations to drop out

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

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ahmad Malik a écrit :
> Hello all
> 
> I am stress testing the rate control algorithms in congested scenarios. 
> We have one AP and ten stations connected to this AP. Now if we use AMRR 
> algorithm on all the stations then the connection to the AP is dropped 
> after a while under moderate to high load.
> 
> The problem seem to be related to stuck beacon problem. My primitive  
> investigation suggests that only during the working of vanilla 
> (non-modified AMRR), the AP side gets Stuck beacon, and restarts the 
> node. This does not happen with Sample or minstrel algorithm. For our 
> work we need to compare AMRR and this stuck becon problem is severely 
> limiting our progress.
> 
> I will greatly appreciate if someone can help me investigate this 
> further or throw in extra ideas for the reasons why it might be happening.
> 
> I am using madwifi version 0.9.4 (latest stable release). In addition 
> wifi mac 10.5, phy 6.1, radio 6.3 and Atheros 5212 version are used on 
> the stations. No encryption is allowed and all Atheros specific 
> enhancements (ff, compression etc.) are turned off. The station are 
> running fedora core 6 (linux 2.6.22 kernel).
> 
> I can give someone access to my testbed consisting of 16 Atheros enabled 
> wireless nodes, if interested. I have also tested it on orbit-lab 
> testbed and it is giving the same problem there.
> 

Hi Ahmad,

This is very interesting. I am not sure this is a stuck beacon problem.
I've done a similar test with fewer nodes and it happens that STA
connection are dropped since they are not receiving beacons.

In order to narrow your problem, here are few questions:
- - could you narrow why STA connection are dropped? Is it initiated by
the STA or the AP?
- - all your STA and AP are running amrr? Or is there a mix?
- - what are the direction of the stream : AP to STA or STA to AP?
- - using one node in monitor mode should give more details on what's
going on?
- - any kernel logs?

In my own experiment, I increased "bmiss" a lot to avoid this problem
(and I was using minstrel, not amrr).

Regards,
Benoit

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFJEZpcOR6EySwP7oIRAvzuAKDizLayt09x4DkC9JMec+FwL1UEZQCfeUEe
LKhDHUJsqIwCIGODrhAFv5E=
=Sw6z
-----END PGP SIGNATURE-----