Re: no wireless networks detected; slow to connect when forced

[Dan Williams <dcbw redhat com>]

On Thu, 2005-01-20 at 00:51 -0600, Ben Liblit wrote:
> NetworkManager fails to detect any wireless networks on its own, even 
> though there are several well within range that do advertise their 
> SSIDs.  When I use "Other Wireless Networks..." to explicitly tell 
> NetworkManager about my own wireless network, it takes a full minute to 
> associate and get an IP address.

That's an artifact of the way that NM treats "Other networks" right now,
which is basically to attempt to verify that the network does exist
before trying to use it.  It essentially means that the "does this
wireless network exist" question is answered twice, I'm trying to fix
that and will commit code soon.  The problem here is Cisco cards, which
don't actually report blank-essid Access Points in their scans at all,
making it hard to determine whether or not a certain access point is
encrypted, which we need to know to make a good choice in how to connect
to that AP.  But not everyone uses Cisco cards, and its a hack that
degrades performance for everyone except to 10% of people that might
actually use Cisco cards.

> Once I have forced a connection like this, my wireless network's SSID 
> ("Luminiferous") does appear in the NetworkManager panel applet's menu. 
>   However, it shows an empty bar for signal strength, suggesting that 
> NetworkManager is trusting me when I say there is a network but doesn't 
> really see the signal on its own.

Note that Linux wireless drivers suck, and not all drivers reliably
report signal strength.  There's quite a wide variety in what drivers
report to user applications, for example Prism54 and Atheros cards
report a percentage value of strength (0 - 100%), while others like
Cisco and Atmel cards report a dBm value (decibels from -128 - 128).
Some cards report values in these ranges that can't accurately be used
to produce a percentage value, because they don't accurately report the
Noise level the card is experiencing, making it clearly impossible to do
Signal-to-Noise ratio calculation and come up with a percentage value.
This is an area that needs Kernel driver work, but at least it doesn't
affect operation of the cards themselves, like other stuff.

> The laptop's wireless interface is an "Intersil Corporation Prism 2.5 
> Wavelan chipset" which has always worked just fine using the orinoco_pci 
> kernel module.  I'm using Fedora's prebuilt kernel-2.6.10-1.741_FC3 as 
> well as NetworkManager-0.3.3-1.cvs20050112.1.fc3.

The kernel wireless driver for Orinoco/Prism/Intersil chipsets is
actually crap right now.  It hasn't been merged from upstream in a
while, and doesn't support wireless scanning at all.  Upstream does, but
it looks like you've tried that and it didn't compile.  Dave Gibson
(Orinoco maintainer) is slowly merging in the necessary changes to bring
the in-kernel driver up-to-date, but scanning support still isn't in
2.6.10.  NM _really_ needs scanning support to work well.

In short, you really need to use drivers that support scanning, which at
this point is the upstream Orinoco drivers (that don't seem to compile
for you, I have only compiled them with kernel <= 2.6.8) or the 'hostap'
drivers, which aren't in the kernel either.  I'm changing around some of
the connection code to make connections faster from "Other networks",
and I've kind of ignored Orinoco cards for a while because (a) they
don't support scanning, and (b) my Lucent WaveLAN Gold card has other
firmware problems (despite being fully up-to-date from the website) that
make the card not actually associate with the AP I've told it to.  (b)
doesn't seem to be your problem though, which is good.  I'll try to see
what I can do about the time issue though.

Dan