This is the environment:
I'm using an AR5416 card acting as an AP with hostap in 11g mode. I'm running g mode because I did not have time to test the 11n work yet. The box is running CURRENT/AMD64.
These are the symptoms:
A few days ago I began to see very big delays between the AP and my laptop (which has an AR9287 with FreeBSD/AMD64 CURRENT). I tried a simple icmp with 1 second between pings. It is quite curious thing that the delays increase from 30ms until around 100ms (those times are in the same order of magnitude of a ping to google) and then they fall til 30ms again. It did not work like that.
I tested with different clients (an iPhone, a laptop running FreeBSD with an Intel 3945, an asus eeePC laptop with windows7 and other devices). All of them showed the same behavior. No matter if I'm accessing the HostAP box or trying to access internet, everything takes long.
The bisect lead me to commit r227364, which is one quite large. Running the previous commit fixes it.
How-To-Repeat: Compile a custom kernel with r227364 in a box with an AR5416 and use the box as an AP.
Further tests show that when an iPhone associates with the AP, the
problem appears. As soon as the iPhone deassociates the latency goes down.
Some tests where made. The tree elements during the tests were: the AP
Box, an iPhone and a FreeBSD 10/CURRENT laptop. The laptop was sending
icmp messages to the AP while the iPhone was associated.
The AP Box has the 192.168.1.100/24 address. The
A first test consisted of sending 10 icmp messages per second from the
laptop to the AP Box. The command used was:
ping -c 0.1 192.168.1.100
When doing this test the latency was constantly growing untill the
time taken to reach the box and back got up to around 110ms, then it
fell to around 10ms (which quite big, I would say three orders of
magnitude.I would have expected about 0.5ms). The frequency of the
phenomenon was of around 80 icmp messages (that is about 8 seconds of
A second test consisted of sending 100 icmp messages per second to
the box. Now the behavior was different. The latency keep falling to
around 20ms and then it suddenly grew to around 110ms. The frequency
this time was of about 8 icmp messages.
I can obtain traces and upload them somewhere and share the links.
For bugs that match the following
- Status Is In progress
- Untouched since 2018-01-01.
- Affects Base System OR Documentation
Reset to open status.
I did a quick pass but if you are getting this email it might be worthwhile to double check to see if this bug ought to be closed.