Poll stations from AP transparently

[Jacques Caron]

At 02:58 PM 10/22/2002, Victor Aleo wrote:
>>Alternatively, if the clients are using active scanning, the neighboring 
>>APs could have a cache of recently seen probe requests. But again, that 
>>won't work with clients using passive scanning only.
>
>Yes, and there is still another problem with this solution: the active 
>scanning is not performed periodically. Therefore, it can happen that the 
>potentially APs to associate changes (in function of the radio quality, 
>multipath, etc.) and therefore the information stored at the AP will not 
>be anymore representative.

That's why I said "recently seen" :-)

One solution might be to listen for all traffic received by the AP, even 
from stations not associated. But I believe that means having the card in 
monitor mode, and on the same channel as the neighboring AP(s). In short, 
that means having several cards in each AP, only one being used in AP mode, 
the others being in monitor mode. Then each AP can know which stations it 
can "hear" (which doesn't necessarily mean the station can "hear" the AP, 
though, but with proper thresholds on signal levels it should be possible 
to get something consistent).

>>A slightly different question: why do you want your AP to reject clients? 
>>A scalability issue to achieve some form of load-balancing?
>
>Rejecting stations from the AP is a mechanism to distribute the load (to 
>achieve load balanced scenario in a WLAN network). For instance, if an AP 
>"knows" that there are other APs in the same ESS less loaded, it will 
>reject  a station  in order to achieve a better performance for each user 
>and for the network.

When you say "reject", do you mean at initial association, or later on 
(forced disassociation)?

If the stations are more or less stationnary, you only care about initial 
association, and you can be sure all stations are using active polling 
rather than passive, then all APs should get the probe request, after which 
you can decide which one should accept the association (or even better, 
which one should send back a probe response, though I don't believe this 
would be possible with hostap mode).

If they are using passive scanning, you might try the following approach, 
though I'm not sure it will work with all clients: when a station tries to 
associate to a given AP, refuse. The station should (?) try another one, 
which should refuse also. Once it has tried all APs (and you thus have 
information about all APs it can reach, along with signal level etc.), I 
guess it should start again (maybe after some timeout?) and then you can 
pick which AP will accept the association. Of course if the station is "too 
intelligent" and caches those APs which rejected the association and never 
tries those again you have a problem...

Other than that, there might be other solutions for the AP to poll the 
stations, though I have to admit I don't know very well all the low-level 
mechanisms like RTS, CTS, ACK, CF-Poll and so on, but others on the list 
probably do :-)

One thing that could help is to know what the scenario is exactly: public 
or private use, pre-determined set of stations or not (i.e. standardized 
enterprise-provided laptops or random student-owned ones), ability to 
impose some settings or not, stationnary or moving, etc.

Jacques.


-- Jacques Caron, IP Sector Technologies
    Join the discussion on public WLAN open global roaming:
    http://lists.ipsector.com/listinfo/openroaming