[ILUG] Delay between link up and forwarding frames?

[Ronan Mullally]

On Sat, 10 Jan 2009, Gareth 'bigbro' Eason wrote:

> 	On the contrary, you seem to have exactly the right understanding.
> Spanning-tree is a loop-avoidance technology, and since computers
> typically only have one connection to the network, it's unlikely they
> can cause a loop. Hence, the 'portfast' directive on Cisco to state that
> you are plugging a single, end-point device into a port, so the switch
> should not worry about spanning tree for that port. (In the case of
> servers with more than one interface, it's at least likely that traffic
> will not be switched between the two NICs, so portfast is probably still
> safe.)

To be a little pedantic, spanning tree is only an issue at layer 2.  You
can safely plug a layer 3 device (server, router), regardless of the
number of network interfaces it has, into a switch port with STP disabled.
Your Linux box only needs to worry about it if it's acting as a bridge.
Likewise, a switch port that's configured to only route traffic doesn't
need to run STP.

Layer 2 devices (switches, hubs, bridges) are the only devices that need
to run STP.  Once you understand how it works and what it will do to your
layer 2 topology it is possible to (and in some instances positively
recommended that you) disable it on some ports.  In such instances however
the onus is on you to get it right.  Once you start messing with STP
you're in "here be dragons" territory.

A word of wisdom learned the hard way.  When you're buggering about with
spanning tree, or if you've got a problem you think it's causing, patience
is a virtue.  The few moments it takes for STP to converge feels like an
eternity.  Go put the kettle on, or sit on your hands and wait a full
minute or two.  I've confused myself more than once by not giving STP long
enough to sort itself out.  This is thankfully less of an issue if your
kit can run 802.1w Rapid Spanning Tree which will figure things out in
seconds.


-Ronan