Port-alpha archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: ifconfig delay on up

>>>>> "dm" == der Mouse <mouse%Rodents-Montreal.ORG@localhost> writes:

    dm> Sometimes, disabling it _is_ the right answer


    dm> part of being a good netadmin is knowing when something like
    dm> that _is_ a right answer.

sounds plausible only so long as it stays vague.

Switched networks should be running an stp, period.  Before when we
used half-duplex networks, if you made a loop in a network of
repeaters most repeaters had the smarts to turn off one of the ports.

There may be times when it's right to buy some cheap piece of junk L2
switch with chinese firmware that hasn't implemented spanning tree.
Or it may be that you've bought hardware with a spanning tree
implementation so broken that it's better to turn it off---but in that
case, usually there is no knob to turn it off in the first place.

    dm> a home network where it'll be a cold day in hell before
    dm> there's ever a second switch

and where you don't care if there's a storm because you're the only
one using the network.  Yes, sometimes you do not NEED an stp.  I
understand that.  It's not the same to say it's not always needed, as
to say sometimes it's right to turn it off.

If you _have_ working STP, I claim that it's never right to turn it

I am not saying OMG OMG it is *DANGEROUS* to turn stp off.  I'm saying
it's never the right answer.

You should know that 802.1w RSTP has some state machine built into the
standard for automatically detecting edge ports and not forcing them
to suffer the 30 second delay---if you have chinese firmware that
implements the new spanning tree standard literally and obliviously,
you probably have one of these and have a ~working STP without the
delay automatically.  You should know big switch vendors like Cisco
and Extreme pervert the state machine described in the RSTP standard
into something more ``conservative'' based on their old manual
edgeport configuration mechanisms from the 802.1d days, so you have to
mark edgeports in the configuration even if you are using 802.1w, but
it might be somewhat safer to mark them now than it was with .1d.  And
you should know that almost any switch which has a knob for disabling
spanning tree (not all do) will have a second knob right next to it
for eliminating the 30-second link-up delay without disabling spanning
tree, so there is no reason to use the first knob.

You should understand that anyone feeding you dire warnings about
portfast is silly and hysterical because it's always possible to make
a temporary loop in a switched network like this:


by connecting the two hubs last, after any 30s delay has expired.  And
this is actually a common use-case where the hubs (or non-STP
crapswitches) are inside two neighboring cubes, and the cubeworkers
get confused by their spaghetti wiring and connect their two
switchhubs together.  Also VoIP phones with integrated 3-port
switches might do it, sometimes.

but STP is still useful in this case because the control plane will,
barring chain of accidents, still notice the loop within a minute and
break it, which is one good reason (not the only good reason) not to
turn STP off but instead use portfast or edgeport mode or whatever.

    dm> And not all netadmins would do that

I'm the netadmin among other things, and I don't take away people's
crappy hubs and don't have a spinning klaxxon in my office that lights
up when someone plugs in a hub, but only because I haven't gotten
around to it yet.  And I'm not letting people keep their piece of junk
switchhubs because I think doing this is the right way to be the
absolute best possible olympian superman of netadmins.  I'm doing it
because <bllaaargh>.

    dm> if you know exactly what ST is and when you should
    dm> and shouldn't turn it off, you don't need any of this
    dm> advice.

IMHO anyone who thinks sometimes it is right to turn spanning tree off, 
needs this advice.

Attachment: pgpgTHJi4kw6p.pgp
Description: PGP signature

Home | Main Index | Thread Index | Old Index