Further improvements to the software scan implementation
Hi,
as already discussed with Johannes we can further optimize the scan implementation by allowing (on an active channel) to send probes as soon as any other frame is received (and updated the NAV) instead of ever waiting 30ms.
Another optimization I thought of while reading [1] would be to take the same approach as Intel in their ucode. If a frame is received while scanning a passive channel switch to active scan mode and send out probes. That should allow us to shorten the time needed to stay on that channel. Luis, do you think this is ok in regard to regulatory restrictions?
Thanks,
Helmut
Quoted 6 lines · Helmut Schaa
Hi,
as already discussed with Johannes we can further optimize the scan implementation by allowing (on an active channel) to send probes as soon as any other frame is received (and updated the NAV) instead of ever waiting 30ms.
There's even two ways of implementing this:
1) the easy way -- just check for received frames in mac80211
2) the good way -- ask the driver to notify us of _any_ received frames or poll the driver or something like that
1) is clearly very easy, but if it's seeing e.g. a control frame or a frame for a different BSS then the device might not pass it up.
But to be honest, I wouldn't worry about 2) and let Jouni implement it if he cares :) And if we want to make this improvement, 1) is good for all hardware anyway.
johannes
Quoted 5 lines · Helmut Schaa
Another optimization I thought of while reading [1] would be to take the same approach as Intel in their ucode. If a frame is received while scanning a passive channel switch to active scan mode and send out probes. That should allow us to shorten the time needed to stay on that channel. Luis, do you think this is ok in regard to regulatory restrictions?
One problem I could see with that is the TX power to use.
johannes
Adding David Quan and Michael Green, my reply in-line below.
Quoted 7 lines · Helmut Schaa
Hi,
as already discussed with Johannes we can further optimize the scan implementation by allowing (on an active channel) to send probes as soon as any other frame is received (and updated the NAV) instead of ever waiting 30ms.
Quoted 5 lines
Another optimization I thought of while reading [1] would be to take the same approach as Intel in their ucode. If a frame is received while scanning a passive channel switch to active scan mode and send out probes. That should allow us to shorten the time needed to stay on that channel. Luis, do you think this is ok in regard to regulatory restrictions?
Good question. We already enable active scanning if a beacon is received from an AP and are world roaming. We do this in mac80211 through a beacon regulatory hint to the wireless core -- regulatory_hint_found_beacon(). The assumption of the beacon regulatory hint is that APs *must* be compliant, and some cards world roam therefore only allowing passive scan on some channels, when you receive a beacon from an AP on a non-DFS channel or channel 12-14 it is safe to assume you can actively scan and beacon on that same channel. Reason for the passive scan flag to exist is for a way to enable some cards to world roam on some channels. Because all APs must beacon and since we will lift this passive scan flag during an initial scan I am inclined to believe we shouldn't bother with processing other 802.11 frames. So in summary I'd suggest to not do this because:
1) We already have the beacons during an initial scan
2) We'd have to figure out a way to ensure the frame was indeed from an AP
Luis
Quoted 5 lines
Thanks,
Helmut