Slow connect times?

[Ben Greear]

On 11/05/2014 10:58 AM, Avery Pennarun wrote:
> On Wed, Nov 5, 2014 at 9:39 AM, Ben Greear <greearb at candelatech.com> wrote:
>> On 11/03/2014 05:35 PM, Avery Pennarun wrote:
>>> On Mon, Nov 3, 2014 at 8:04 PM, Ben Greear <greearb at candelatech.com> wrote:
>>>> If I am reading my results right, it takes over 2 seconds to associate an
>>>> ath10k station with WPA2 PSK on my systems.
>>>>
>>>> I was thinking this should be quite a bit faster than this...
>>>>
>>>> Has anyone done any similar measurements?
>>>
>>> I haven't noticed any delay like this.  WPA2 PSK negotiation is on the
>>> order of ~200ms or less. Then DHCP screws around for a few seconds,
>>> often.
>>>
>>> A packet trace would probably be useful for narrowing it down :)
>>
>> First part of my problem is that I had some extra debugging turned on and
>> that was being sent to serial console, which is slow.
>>
>> But, even with that off I am still seeing ~200ms at best.
>>
>> Ath9k system on crappy old Atom processors usually completes in
>> less than 20ms for comparison.
>>
>> Off to go spelunking in the ath10k code :P
> 
> That's very interesting for me.  200ms was kind of a worst case in my
> traces, and I'm mostly seeing 20-40ms.  I think the key exchange is
> all hostapd, so I'm surprised it would take so long.  Do you have a
> trace?
> 
> Note that I *have* seen weird problems with delayed probe packets to
> some clients, which could cause things like this.  But that's
> apparently only because I applied Michal's patch to get rid of queue
> blocking :)
> 

My first station comes up in ~20ms, the second one (on same machine) takes ~200ms.  If I try
bringing up 10 as fast as supplicant can do it's thing, then it takes around 1 sec for each
station.

Does not appear to be an AP problem, as ath9k stations to the ath10k AP are all
in the 20ms range.

I'm updating my test systems Fedora 20 to have easier use of some debugging tools and will then
start figuring out where the slowness is.  I'm running my firmware with patches that
should help with flush related problems, but that is still my first suspect!

Thanks,
Ben

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