Re: NetworkManager and STP

[Peter Jones <pjones redhat com>]

On Wed, 2006-03-22 at 13:14 -0500, Darren Albers wrote:
> > The way this is supposed to work is that you configure STP for each of
> > the ports you're using to communicate with other switches, and thus you
> > get high availability out of it.  What actually happens is that network
> > admins get lazy, and instead of planning their network, they turn STP on
> > for *every* port, so they can have a randomly-connected network.  And
> > then everybody has to play the STP game, not just bridging devices.
> 
> That isn't completely true, at least in the Cisco world.  Cisco has
> STP on by default on every port (Which to me is a proper decision) and
> it should never be disabled.  To resolve the issue Port-fast should be
> enabled on Access Layer ports.

Portfast doesn't actually solve the whole problem either -- that's why
they invented "bpdu-guard" later, which is a somewhat better approach.
But that only works for cisco -- there are still other vendors on the
market implementing STP, and some of them don't have anything
portfast-like at all.  So the client has to be aware of STP, or it has
to punish others with absurdly long timeouts.

Of course, since the client basically has to be involved anyway, the
*real* answer would be to fix the protocol so that a non-bridging device
can NAK BPDUs.  Right now the closest you can do to it is to claim
you're a switch with 0 bridge ports.  I haven't tried it (yet), but I
bet it's a great way to figure out which switches implement the
re-balance criteria incorrectly, and will turn out to be a wonderful way
to crash whole networks at once.

> So like you pointed out in a properly architected enviroment this is
> not an issue since a laptop is being connected to an access layer
> switch which should have port fast enabled (or STP disabled) and the
> device taps in quickly.
> 
> Originally I was thinking that this something that should be resolved
> by NM, but now that I think about it I wonder if Network Manager
> should concern itself with an improperly architected network?

I'd try to argue no, but I'm 100% sure we (at least RH, maybe upstream
GNOME too) will eventually see an unending torrent of bugs filed about
it.  We can probably at least put it off for a while, though.

-- 
  Peter