Brian E Carpenter <[EMAIL PROTECTED]> wrote:
|Dan Lanciani wrote:
|>
|> "Michel Py" <[EMAIL PROTECTED]> wrote:
|...
|> |One billion routes in the global routing table = does not scale.
|>
|> This is the main fallacy in your statement. You are assuming that a billion
|> PI address blocks has to equate to a billion routes in some global routing
|> table (or even that there has to *be* a global table). That would be the case
|> only if we insist on remaining with a full-knowledge centralized routing model.
|> Such models are outdated. We can do better.
|
|We tried (NIMROD) to develop alternatives to full-knowledge flat
|routing; but that attempt failed.
I'm not familiar with the specific constraints placed on the attempted solution,
so I can't say whether it really had a chance. Nevertheless, I wouldn't read
too much into the failure. There are clearly distributed approaches that will
work. They have their various tradeoffs, advantages, and disadvantages. As
I've pointed out over and over through the years, we really need to decide up
front what cost we are willing to pay for a solution. As long a significant
number of people argue that the cost must be zero because the benefit is worth
zero or is even negative, any solution will be deemed a failure.
|However, as Noel would have told us
|for the N'th time if he was watching this thread, any such alternative
|requires a method of abstraction and summarization of routes, a.k.a.
|aggregation.
No, this is exactly the fallacy that leads to failure. There is no need
for summarization of routes because there is no need for any given node
to have a complete picture (even summary) of the topology. The whole point
of distributing the routing task is to make the resources available for
routing grow in aggregate at least as quickly as the consumers of those
resources. As soon as you start looking at summaries of the whole table
you focus resource usage for the whole net on individual nodes again.
|Today, the only way we know how to do that is
|by shorter and shorter prefixes on binary numbers.
And that's the trap. As long a you start with the false requirement for
summarization you end up back at aggregation.
But we've been over and over this through the years and it always comes down
to the same hand-waving arguments that "it can't be done." Exposing the
specific assumptions behind those arguments is like pulling teeth, and
ultimately any proposal can be shot down with the cost-must-be-zero argument.
It's starting to get really boring, IMH("historically inaccurate")O.
Dan Lanciani
ddl@danlan.*com
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
