Re: Please don't use 11.1-MFIB with DVMRP-mode tunnels

[Göran Bengtson <]

On Wed, 1 Apr 1998, John M. Zwiebel wrote:

> From: "John M. Zwiebel" jzwiebel@localhost
> To: Göran Bengtson goeran@localhost
> Cc: David Meyer dmm@localhost, fenner@localhost,
>     mbone-eu-op@localhost, mbone@localhost
> Message-ID: <199804020033.QAA06512@localhost>
> Date: Wed, 01 Apr 1998 16:33:26 -0800
> Subject: Re: Please don't use 11.1-MFIB with DVMRP-mode tunnels 
> 
>  ^ 
>  ^ I've seen the same problem for 11.3(1 and 2) and it applies to both
>  ^ tunnels and DVMRP-neighbors on LANs. On LANs the work-around may not
>  ^ be feasible since 'dvmrp unicast-routing' may be needed for RPF-checks.
>  ^ 
> 
> "ip dvmrp unicast-routing"  should _never_ be configured on a DVMRP tunnel
> under any circumstances.  never ever.

I agree (DVMRP-tunnels and DVMRP-based RPF-checks even worked before the
'unicast-routing' command was implemented, if I remember correct) - but:

1	This is not very clear from the Documentation.

2	It doesn't change the fact that I've seen the problem with
	a LAN-interface never showing up in the Olist for (S,G) if the
	downstream DVMRP-neighbor (the only one) on that LAN poisoned the
	route (or an aggreagate of the route). It clearly has to do with
	the changes made to use poisoned-reverse from downstream DVMRP-
	neighbors ("conditional flooding"). I've seen it in 11.3(1 and 2).
	I don't know if it present in the 11.1-MFIB too.

	In the case of DVMRP-neighbors on LAN, the 'ip dvmrp unicast-routing'
	is needed.


			/ Göran Bengtson
			  Chalmers Univ. of Technology