Re: [Idr] draft-jakma-mrai-dep-00 (fwd)

On Nov 25, 2008, at 11:59 PM, Olivier Bonaventure wrote:

> Pedro,
>>
>> In general my recommendation for MRAI values would be 0 for eBGP  
>> and 0
>> for iBGP. As the draft correctly explains calculating a value that  
>> best
>> fits the network is a very tricky business.
>
> I would not recommend using the same MRAI timer for iBGP and eBGP.  
> In a
> large AS using route reflectors or confederations, we are often in a
> situation where some routers of the AS learned an alternate path but  
> did
> not distribute it widely due to the alternate paths being hidden by
> border routers

How does applying MRAI help in this case ? And how does having an  
higher value for eBGP rather than iBGP help ?

It seems that in your scenario the faster the route propagation the  
better.

> (e.g. different local-pref or longer AS-Path), router
> reflectors or at the borders of confederations. See e.g.
> http://inl.info.ucl.ac.be/publications/quantifying-bgp-routes-diversity-insi
> but discussions with operators showed that this is true for other  
> networks.
>
> In such large ASes, when a link fails iBGP should converge first in
> order to allow all routers to learn the alternate path and advertise  
> the
> alternate path over eBGP sessions.

What if your alternate path happens to be being an eBGP peer ?

Dest prefix  is source by AS A

AS A connects to AS B preferentially and has fallback routing though  
AS C. Where C only advertises destination prefix if it is not  
otherwise know. AS D is transit for both B and C.

Convergence requires D to send withdrawal to C to receive the  
alternate path back.

It is not unheard off to have such arrangements.

Please do keep in mind that i didn't suggest that the values (0, 0)  
are the best for your network and for the scenario that you are trying  
to optimize. I'm merely suggesting that as general values for the  
generic case (0, 0) are hard to beat... and they work just fine.

> This can be achieved, at least
> partially, by using a longer MRAI timer for eBGP compared to iBGP.
> As you note, the situation is different in BGP/MPLS VPNs  
> environments or
> when draft-walton-bgp-add-paths-06.txt is used.
>
> I would not personally recommend to exempt withdraws from the MRAI on
> eBGP sessions in large ASes.
>
>
> Olivier
>
> -- 
> http://inl.info.ucl.ac.be , UCLouvain, Belgium
> _______________________________________________
> Idr mailing list
> Idr@ietf.org
> https://www.ietf.org/mailman/listinfo/idr

_______________________________________________
Idr mailing list
Idr@ietf.org
https://www.ietf.org/mailman/listinfo/idr