[quagga-dev 5445] Re: [PATCH] RFC 2328, chap 8.1:

Andrew J. Schorr aschorr at telemetry-investments.com
Tue Jun 3 16:38:43 BST 2008


On Tue, Jun 03, 2008 at 04:03:13PM +0100, paul at clubi.ie wrote:
> a) Whether zebra should detect shared addresses automatically or not
>    (the alternative is that the administrator needs to explicitly
>     mark the shared addresses)
> 
> b) Whether unnumbered addresses should be *administratively* enabled
>    by default (ie automatically or not)
> 
>    corrolary: Whether it should be enabled globally, or per interface
> 
> There's definite disagreement on b - but it's a trivial detail that 
> shouldn't be a barrier to anyone wanting to add unnumbered support 
> ;).
> 
> I think 'a' is a requirement. If we don't automatically recognise 
> shared addresses then we setup users for failure ("you mean I had to 
> set some flag because I assigned an address multiple times? Oh..").

Currently, with the existing code base, it is fine to have shared
addresses (i.e. same local address used on multiple interfaces), as
long as the PtP interfaces have unique peer addresses.  This works
fine currently (case 'a'), so I see no problem.  Why is 'a' a requirement
if it is working fine today with existing code without this detection?

But if by "shared addresses" you mean PtP links with both the same local
and peer addresses, then yes, that is a new situation not currently
supported, and auto-detection might be nice.

> I really don't see the attraction in forcing the admin to configure 
> that manually (this is 'a' above, not 'b'). Routing is already 
> complicated enough, why do we have to make it more complicated than 
> needed?

I don't see why 'a' requires any configuration at all.  Most likely I'm 
confused.

As for 'b', I still tend to think this should be enabled administratively, but
the easiest/best way (in my opinion) would be to specify that all PtP
interfaces with matching local and peer addresses should use unnumbered
behavior.  Something like:

   router ospf
     # configure unnumbered behavior on all PtP interfaces with
     # local address <localaddr> and peer address <peeraddr>
     unnumbered <localaddr> <peeraddr>

But maybe I'm just thinking of this from a linux perspective that is
not appropriate for other platforms...

Regards,
Andy



More information about the Quagga-dev mailing list