Periodic scanning and larger beacon timers.

[Ben Greear]

On 01/25/2011 08:18 AM, Jouni Malinen wrote:
> On Mon, Jan 24, 2011 at 04:11:41PM -0800, Ben Greear wrote:
>> I'm currently using an AP with beacon timer of 240.  I believe this is
>> beyond the normal dwell time for passive scans, and what I'm seeing is
>> that occassionally, wpa_supplicant gets into a state where it scans every
>> 5 seconds but doesn't ever find the AP.
>
> Which channel are you using? With active scanning, it there should be no
> problems in finding the AP. However, with passive scanning, such a long
> beacon interval can certainly cause problems.

Channel 6 (2.4 Ghz range).  I don't know for certain that it uses
active or passive scanning.

>> Any opinions on adding some randomization to the scan timers so that
>> they cannot get stuck in a dead zone between beacons?
>
> Which scan timer are you talking about? wpa_supplicant does not
> currently control the dwell time or exact timing of channel changes
> during the scanning operation. As such, this would likely apply more to
> the driver/kernel 802.11 stack implementation than wpa_supplicant.
> Changing the about five seconds timer in wpa_supplicant is unlikely to
> have noticeable difference since it is not exact five seconds between
> each scan, but five seconds after the previous scan was completed and I
> would fine it surprising if you can synchronize this with any AP beacon
> operation (those are in TUs, not milliseconds and for example, that 240
> TUs would be 245.76 ms).. Anyway, if you can show that changing this
> would really have a noticeable effect in real life, I would have no
> problems adding random 0..999999 usec extra delay to that wait period..


First, I was testing some scan optimizations in mac80211 (don't go off
channel when scanning only the current operating channel).  It's possible
I have bugs in this code, though it seems to work fine at least most
of the time.

Another thing:  Since I was scanning only a single channel, I think the
timing between start and stop of a scan would be pretty regular, which
might exacerbate any periodic scan interval issues.

And, I didn't wait too long before initiating a scan manually, so it's
possible the scan would have started working again soon.

All that said, I think it's still a valid theoretical problem to have
periodic scans on very regular intervals.  I'll work on a patch to
add some randomness to at least that 5-second timer and add it to
my ever increasing patch queue :)

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com