[c-nsp] Default-Information Originate supplimental question

[Peter Rathlev]

Hi Michael,

I'm not too strong on redistributing, but have some general comments
regarding the BGP part. I hope someone else will correct me if I'm wrong
here.

On Fri, 2008-07-04 at 15:39 +0100, Michael Robson wrote:
> Guys, thanks for the clarification; but now further questions leading  
> on from this. My understanding is that by default (without overriding)  
> if you are redistributing routes from BGP into OSPF on the router and  
> some routes  have been learned via iBGP, then these iBGP-learned  
> routes will not be advertised since it is an IGP->IGP exchange (which  
> might cause loops). Following from this, since as I understand it,   
> "the default-information originate" command is a special case variant  
> of the redistribute command,

The "default-information originate" does not redistribute anything in
itself. Focus on the "originate" part -- if this BGP speaker knows a
"default" (not from BGP) and the default is lifted into BGP (via
"network 0.0.0.0" or some kind of redistribution) the BGP speaker will
only actually tell its neighbors about it if it can "originate" the
default.

>                              then the default route, by default,  will  
> not be injected into OSPF if it had been learned via iBGP - correct?  

Your redistribution is not related to whether or not the box originates
a default. In order for BGP to originate a default, it has to have the
default from somewhere else, like an IGP or a static route. The
"default-information originate" just allows a 0/0 route to be announced
via BGP, which by default will not happen.

If a box learns the default via BGP (e.g. from a BGP neighbor with
"default-information originate") and you unconditionally redistribute
everything to some OSPF process, I think this process will get the
default, "originate" or not. And a BGP speaker will relay a default also
without the "originate" command, since it doesn't _originate_ the route.

> If this is the case, then if you have the situation on a router where  
> there are 2 default routes being learned, one via iBGP and one via  
> eBGP and the _iBGP_ route is preferred over the eBGP (e.g. it has a  
> lower MED value), then would the lesser preferred route be injected  
> since the other cannot be by default, or would neither be injected  
> because the preferred default route has been learned via iBGP?

That may be a good question, but it sounds like a dangerous thing to
try. :-) Do you need this redistribuion from BGP to your IGP? Are you
using synchronization?

> Finally (phew), can anyone give a possible explanation as to why none  
> of our eBGP-learned routes have an origin type of e (i.e. they are all  
> of type i)?

Origin "e" is "EGP" a legacy protocol (probably) not used anymore. So
either you have "?" (incomplete, e.g. redistributed routes) og "i" (IGP,
e.g. "network" statements).

Regards,
Peter