From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Wed Feb 26 2003 - 14:27:20 GMT-3
At 3:17 PM -0400 2/25/03, Joe Chang wrote:
>The original question is whether split horizon is used to prevent loops in
>BGP.
>
>Short answer is no, BGP will not send a route back to the advertising AS
>because BGP is aware that route has traversed that AS. This feature can be
>implemented regardless of where the routing protocol sits on the OSI.
_can_ be implemented. In other words, an implementer choice, which
may or may not be user-configurable. Depending on what you are
trying to do in the implementation, it may make sense to send out
garbage that you know will be rejected at the destination, if it
significantly saves processing resources at the source and bandwidth
is cheap.
This is not the trend, however, in either BGP architecture or
implementation, as evidenced by things like ORF. Maximum
preprocessing of updates is particularly important in BGP because
avoiding sending irrelevant or incorrect information contributes to
Internet stability.
Understanding the goal of stability is important to several current
BGP threads. I think it bothers some people, for example, that you
don't always get completely optimal route selection in a RR cluster.
Giving up route optimality for stability is a design choice.
>
>Howard, we don't care if you personally invented BGP, just give it to us
>straight, please?
What, Sir, do you mean by "straight"?
I personally rather like to know someone's credentials for making a
flat statement, be those literature references or personal experience.
This archive was generated by hypermail 2.1.4 : Sat Mar 01 2003 - 11:06:36 GMT-3