Hi,

Feedback on suggested path changes from another operator:

I imagine that in our case having *all* route-servers consistently add their 
ASN would not lead to too many path changes. The changes that we will see is 
that we will prefer direct peerings over route server prefixes because of 
AS-Path length (today it is another metric that is the tie breaker, since 
AS-path length is equal). 

For our transit customers however, this may be a slightly different story 
because they suddenly see the direct peerings with a shorter as-path than 
routes propagated via route servers and may select different paths.

Suggestion:

I would prefer a solution that has no impact on the as-path: 

> On 19.12.2017, at 00:47, Jared Mauch <[email protected]> wrote:
> 
> I do think that having a RS emit a community saying “RS_ASN:xxx” would be
> of value.  As mentioned in other emails, finding these edges can be quite
> complex when doing operations.

However, communities are very fragile when it comes to route propagation. 
Removing / modifying communities is happening everywhere (mostly at AS 
borders). I would not like to rely on BGP communities for this.
 
We have a solution to a slightly similar problem in iBGP route-reflection. For 
that purpose we added an optional non-transitive attribute CLUSTER_LIST. I 
suggest a similar approach for IXP route-servers (ROUTESERVER_LIST 
optional,transitive = list of route-server ASNs that have propagated the 
prefix). That seems less fragile and the implementation changes are limited to 
the route-servers themselves.

Christoph

_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to