[PATCH 0/3] hostapd: introduce Automatic Channel Selection (ACS)

[Helmut Schaa]

On Mon, Aug 26, 2013 at 8:32 AM, Michal Kazior <michal.kazior at tieto.com> wrote:
> On 23 August 2013 10:56, Helmut Schaa <helmut.schaa at googlemail.com> wrote:
>> Hi Michal,
>>
>> On Fri, Aug 23, 2013 at 9:19 AM, Michal Kazior <michal.kazior at tieto.com> wrote:
>>> Hi,
>>>
>>> This is an attempt at ressurecting ACS effort for hostapd.
>>>
>>> This is based on the work done by Luis R. Rodriguez
>>> <lrodriguez at qca.qualcomm.com>.
>>>
>>> I've rebased and improved the patchset. Notable changes:
>>>
>>>  * Interference factor calculation has been modified.
>>>
>>>    The formula now includes the relative channel noise floor so a channel with
>>>    lower noise floor should be preferred in case channel load evaluates to 0.
>>>
>>>    The formula now spits out positive values that can be summed up, which is
>>>    needed for:
>>>
>>>  * Overlapping/primary/secondary channels are all used when calculating total
>>>    intereference factor for a given connection type (2.4Ghz, HT40, VHT80).
>>>    This most likely still needs tuning but it's a good start I think.
>>>
>>> RFC v2:
>>>  * remain-on-channel replaced in favour of scan-based survey data gathering
>>>    (hostapd: Add offchannel support -- dropped)
>>>
>>>  * chan_time scan parameter patch is introduced
>>>
>>>  * tested with ath9k and ath10k
>>>
>>> RFC v3:
>>>  * chan_time patch is dropped in favour of doing a couple of passive scans
>>>    this should still allow gathering enough data for ACS
>>>
>>>  * lots of cleanups and some small fixes
>>>
>>>  * trimmed down main ACS patch commit message
>>>    all details are already in code comments or on wiki page [1]
>>>
>>> PATCH v1:
>>>  * rebased on latest master branch
>>>
>>> The following link is a git patchbomb that has all 3 patches in case the
>>> mailing list server decides to spam-block my patchset.
>>>
>>>   http://ix.io/7vF
>>
>> Just gave your patches a quick try (with ath9k).
>>
>> One thing I noticed was that hostapd tries to respond to probe
>> requests while doing the ACS channel scan
>> resulting in a couple of "handle_probe_req: send: Resource temporarily
>> unavailable" errors:
>>
>> 1377247507.112841: Configuration file: /var/run/hostapd-phy1.conf
>> 1377247507.223379: ACS: Automatic channel selection started, this may take a bit
>> handle_probe_req: send: Resource temporarily unavailable
>> handle_probe_req: send: Resource temporarily unavailable
>> handle_probe_req: send: Resource temporarily unavailable
>> handle_probe_req: send: Resource temporarily unavailable
>> 1377247513.287024: Using interface wlan1 with hwaddr 00:03:7f:11:20:08
>> and ssid "OpenWrt"
>
> Hi,
>
> Hmm.. The overlapping BSS scan should probably yield the same issue,
> no?

Good, point, haven't checked that but seems likely.

> Should hostapd ignore all management frames until all BSSes have
> beacon_set == 1? Won't that break anything?
>
> Anyway I think this should be (if at all) dealt with in a separate patch.

Agreed.

Helmut