Scott Huddle wrote:
Interesting reading. Has anyone observed this operationally?
I don't think so because you have to construct a strange topology to achieve the first described effect (persistent loops); and the second effect ("incorrect" route computations) is pretty much what can be a desired behaviour. The definition of iBGP mesh's behaviour as ideal is questionable in many cases. So, yes, route reflector can behave differently but that can actually be a plus. As long as you know what you're doing. --vadim PS. Actually, even #1 is not a drawback. There's a lot of ways to create persistent loops with creative usage of iBGP mesh alone. That only shows that the tool is powerful enough :) All in all, the draft is a good piece on less than obvious dangers of fancy routing techniques.