[Madwifi-users] performance and reliability problems - solutions on the horizon? (madwifi-bsd, hostapd wpa)

[Ethan T. <et...@et...>]

I'm having performance and reliability problems with my madwifi based 
802.11g access point.  I've been researching this but I'm stuck.

I first noticed that the rx rate on my XP client was much lower than the tx 
rate.  In my usual work location (one floor up from the AP), I was receiving 
at 80KBps while I cound send at 220KBps.  I figured out that this was 
because madwifi had decreased it's tx rate, while my XP client was happy to 
transmit faster. (!)  With my old 11b bridge, I had stable 150KBps 
connections at this same location.

Sitting next to the AP, I measured TCP throughput at 2.25 MBps round-trip. 
Which was not bad, as long as it lasted.  Over time the AP's tx rate would 
drop to 9Mbit or lower, and it wouldn't ever recover.

I thought it would be clever to switch from onoe to aamr rate control. 
Initial impressions were good, because the AP's tx rate would quickly 
recover to higher rates, when the client was near the AP.  But, aamr also 
broke the connection when the client moved to the edge of the signal 
(upstairs).  Because aamr effectively reduced the range of the AP, I gave up 
and switched back to onoe.

Occasionaly, the wireless net dissapears altogether.  I've seen this happen 
while I was sitting at the AP, and nothing was revealed by the system logs. 
The interface simply stopped transfering packets for period, then started 
again.  Active socket connections such as ssh died when this happened.

Why all this strange behavior?  Querying with athstats reveals that roughly 
half of the packets sent are 'long on-chip tx retries', and 1/8 of them are 
'tx frames with no ack marked'.  There are also a large number of PHY errors 
(OFDM timing and CCK timing).  This makes me wonder, am I looking at a HAL 
problem or a driver problem?

After digging around in the list archives, I see there are other folks 
having similar experiences with madwifi.  Are these problems known to the 
developers?  Is there a fix in sight?  I am happy to help track these 
problems down, if I can.

Thanks,

Ethan

PS, details about my environment are in this post:
http://article.gmane.org/gmane.linux.drivers.madwifi.user/7373