filtering long prefixes
smd at icp.net
Thu Sep 21 10:50:03 UTC 1995
I agree with you 100% that the Internic had adopted a
policy of handing out longer prefixes than I had initially
warned everyone Sprint would be refusing to accept from
external neighbours. I also note that the Internic also
had wording which pointed out the dangers of accepting
this sort of prefix from them, right on its application
form, which I reproduce below. I think that the InterNIC
did its job of warning people of upcoming realities,
and therefore its policies were not, in fact, in conflict
206/8 was specifically chosen, in fact, because
no allocations had been made from it at the time.
Now, I note that I have pointed out that we are willing to
be flexible when it comes to /19s, but for the moment, the
filter will reject anything longer than /18s in 206/8 to 239/8.
In part this is because it gives me the opportunity
to study who has gotten disconnected and how difficult it
will be to reconnect them without doubling the number
of prefixes we will see in 206/8, not to mention 207/8 to 239/8.
You may call it direct engineering terrorism if you like.
I call it a sound, planned, well-announced step to keep the
maximum growth of the routing tables confined to the level
at which technology exists to support them.
My only (large) regret is that any long prefixes in
206/8 ever worked at all, principally due to an oversight in
a reconfiguration of ICM AS 1800. I hate breaking things
that were already working...
To translate that sentence: "Maybe. Let's see how bad
the mess is first."
Due to technical and implementation constraints on the
Internet routing system and the possibility of routing
overload, certain policies may need to be enforced by the
major transit providers in order to reduce the number of
globally advertised routes. These potential policies may
include setting limits on the size of CIDR prefixes added to
the routing tables, filtering of non-aggregated routes, etc.
Therefore, addresses obtained directly from the InterNIC
(non-provider-based, also known as portable) are not
guaranteed to be routable on the Internet. ^^^^^^^
More information about the NANOG