On 04/10/07 20:51, Stephen J. Gowdy wrote:
> On the CERN network it is unstable but NetworkManager reconnects and my 
> sessions just pause for a few moments. Not perfect but workable. I've no 
> idea what makes the CERN network special.

Neither have I, could be ratio of 11b to 11g in your vicinity.
We tried to backport ipw3945 to SL(c)4, this never has been fully stable 
but seems to work form some models (I have no first-hand experience on 
SL5). Known issues we found were

* long time to associate - the thing would (slowly) scan 11g first, then 
11b. By the time it got down to 11b access points, "ifup" (dhclient) 
sometimes would have decided to give up.

* 1.14.2 firmware seems to improve association speed

* the binary userspace governor daemon (ipw3945d, version 1.7.18-1) can 
get into weird states. We first tried to hook it into the module load 
phase, this was rather unstable. Now this runs as a standard service, 
this still causes some warning message at boot (..delaying 
initialization..) but seems overall much more stable since the driver is 
loaded way before the governor starts

* signal quality as reported by the driver is consistently low.

I also remember having seen the "invalid misc" counter go up over time - 
this could be an indication that we are unable to understand one packet 
type that presumably is required for stability... perhaps sb with the 
hardware could check whether this is indeed the case. The module 
supports a "debug=" bitmask that should make it rather verbose, see 
ipw3945.h for details.


Hope this helps somebody..
Jan