Handoff insights with Prism2 linux driver

[Victor Aleo]

Hi all,

I am studding the handoff time at the MAC layer with 2 APs and 1 
station. All of them are running the Prism2 driver (Prism2-2002-05-1). I 
have measured, analysing the 802.11 frames with Ethereal, the next 
period times when I perform the handoff test (I create a flow data with 
MGEN from the station):

Period of time 1 (from the last received ACK to the first Probe Request 
generated by the station): This time is around 2 seconds, which is 
totally unexpected and inadmissible!

Period of time 2 (from the first Probe Request generated by the station 
to the Reassociation Response from the AP): This time varies from 220 ms 
up to 260 ms, which was the expected value.

Thus, with this results the total handoff time is the sum of the two 
period times: up to 2.260 seconds. It is obvious that there is a problem 
here! Indeed, the problem is within the first period time. I have 
analysed what the station does when it is disassociated. These are the 
results: once the station is disassociated it sends periodically (during 
around 2 seconds) 2 Data frames (1 Data frame and the retransmission) 
and 3 Request-to-send frames (RTS) without receiving ACK frames!. Of 
course, all these Data frames are lost (MGEN use UDP traffic). This goes 
on until the station sends the first Probe Request frame (actually it 
sends up to 3 Probe Request frames, I guess to be sure or to compute an 
average?). Then, the second AP replies with 2 or 3 Probe Responses and 
afterwards it sends the Reassociation Response and the handoff is finished.

I cannot understand what is happening. Thus, why the station takes so 
long to send the first Probe Request? Because this is the reason to 
explain the high value of the handoff time. As far as I know, there 
should be only 7 Data frames without ACK, and in one of the tests there 
were around 100! Is there something wrong with the Prism2 driver for 
linux and the handoff issue?

Sorry for this long email, but I think is a quite interesting issue.

I would appreciate any comment or suggestion. Thanks,

Victor