[ath9k-devel] TX packets dropped

[David Ehrmann <ehrmann@gmail.com>]

On 11/21/2010 4:51 PM, Brian Prodoehl wrote:
> On Sun, Nov 21, 2010 at 2:08 AM, David Ehrmann<ehrmann@gmail.com>  wrote:
>> I have an AR9220 mini PCI card (Sparklan WMIA-198n) on an Alix 3d2 board
>> running OpenWrt (2.6.32.25 kernel).  I set up the radio in AP mode, give
>> it encryption, then connected with another computer with an Intel
>> 802.11n card.  The connection wasn't reliable, so I tested it with
>> iperf.  When the client sends packets, they almost all make it, but when
>> the AP sends them, no more than 66% make it.  I'm trying to figure out
>> what the problem could be.  It shouldn't be interference--this is in a
>> 5ghz band, and RX was fine.  It could be a bad pigtail or antenna, but
>> I'm not sure how I can test for that.  It might be the motherboard's
>> power supply, so I'm looking into how many watts it can supply (the card
>> specs say it takes ~2.5 watts).  I tried forcing 802.11g mode, and the
>> issue was still there, and I dropped TX power, but it didn't seem to
>> affect anything.
>>
>> Any ideas?  Is there a way to check stats on each antenna?
> What do ping times look like between AP and client?  I'm seeing
> similar behavior with OpenWrt trunk (r24043) and an AP with an AR9160
> chip.  I'm having TX traffic queue up on the AP for seconds at a time,
> so I'll have periods of no traffic at all, and then all the pings will
> come through at once, with times of 8s, 7s, 6s, 5s, 4s, 3s, 2s, 1s,
> and 1.97ms or whatever.  I have two other APs here with different
> drivers and chipsets that don't exhibit this.  I don't remember seeing
> this with recent builds of backfire, so maybe its a relatively recent
> regression in OpenWrt.  You're running trunk, right?  What svn rev?

My ping times are typically between 2ms and 5ms, but around 1 in 15 is 
more like 15ms to 50ms.  My 802.11g AP behaves roughly the same.  
Traffic could be bunching up, though.  Sometimes the console (over SSH) 
won't respond when I hit enter, but a few seconds later, all the new 
lines show up.

I'm using r23987, so it might be the same issue.