Re: Why BSSID? (was Re: A couple of minor issues with NM)

[Dan Williams <dcbw redhat com>]

On Tue, 2006-12-26 at 14:18 -0500, Derek Atkins wrote:
> Dan Williams <dcbw redhat com> writes:
> 
> > SSID before.  It's all based of SSID, and while that may have worked in
> > the past, we'll probably have to change that to some combination of SSID
> > +BSSID sooner rather than later.
> 
> I still don't understand why you need this change.  Why is the SSID
> not sufficient?  I've seen so many problems with NetworkManager in
> multi-AP networks.  I've always written this up to NM being too
> fragile w.r.t. multiple APs.
> 
> Could you please explain why the ESSID (and mode, a or b/g) isn't
> sufficient in 99% of the cases?  The only times I can really think
> where you'd need the BSSID is in cases of AdHoc networks, or for
> networks like "linksys".  Is there some non-obvious issue that
> I'm failing to comprehend?

I guess I've been a bit unclear; we need to treat "networks" as an SSID
_plus_ some combination of other properties.  These properties include:

- mode: Ad-Hoc vs. Infrastructure
- band: a vs. b/g vs. a/b/g  (ie, YOU [1])
- encryption: none vs. WEP vs. WPA

Essentially, all of those plus the SSID should describe a network, and
we should not automatically connect to a specific BSSID if it doesn't
conform to the stored attributes described here.  We should also likely
start breaking these different combinations out in the GUI menus, using
different icons or some other means of differentiating same-named SSIDs
that have, say, different encryption options or a different mode.

Right now, it's _only_ SSID-based.

Dan

[1] at MIT with the APs on A-only bands with the same SSID as the APs on
B/G-only bands