Noel,
An engineering problem has been raised, and while a new architecture can
be a solution to an engineering problem, I want to understand just how
far we have to go to fix the engineering problem. In order to do that I
have to understand the engineering problem better. And so I wonder if
someone can point me to the presentations that were given at the routing
area so I can see what we're talking about. Amazingly, in the seven or
so presentations I've taken on this topic, only one of them had any data
at all (thank you Tony & Vince) about routing growth, but even it did
not answer this question: is our primary concern control plane churn in
the form of memory bandwidth or memory size? I've gotten an oral answer
of "both", but at least some data is needed in order to answer the
question of what sort of characteristics any mapping from id to loc must
have. That was the point that I amongst others tried to make at the
mike, today.
Which brings me to Bernard's point:
a. Performance data is often highly dependent on the implementations being
studied, and therefore it is often unclear whether the conclusions hold
universally.
b. There are often multiple solutions available to the problem, some of
which will perform better than others in given use cases. Without
agreement on the use cases, it can therefore be difficult to agree which
solutions are superior, even if the performance data itself is
unassailable (which is itself rare).
Understood, and so if you can tell me what "the problem" is with some
reason to believe so, I believe we will have gotten over this hurdle. I
believe therefore the questions I asked are not beyond the bounds of reason.
Eliot
_______________________________________________
Int-area mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/int-area
