Routing wars pending?

Wed Nov 15 21:59:30 UTC 1995

    Date:        Wed, 15 Nov 1995 15:16:59 -0000
    From:        Sean Doran <smd at icp.net>
    Message-ID:  <95Nov15.151700-0000_est.20701+5 at chops.icp.net>

    What's important is the routing system.  

This is certainly true.

    The fact that you are registered in registry X as being
    the "owner" of prefix Y is completely irrelevant if pretty
    much the entire Internet is going to route X towards Z.

As is this from a practical sense - but we really need to
look a little beyond that.

If we simply look at the routing system, we find that longest
match preference is the way to go, which argues, very strongly
that in order to keep routing working, everyone should inject
(at least) /24's (no shorter) prefixes, so they won't be
overrideden by a longer prefix from elswhere, to be absolutely
safe all advertisments should probably be /32's for every host
in every organisation, backed up by /31's /30's /29's ... going
back as far as needs routing so the longest possible length
will get through any prefix length filters that exist.

I don't think that is quite what cidrd is attempting to
accomplish.

If we step beyond letting the routing system be the sole arbiter
of what addresses are useful, so people can be encouraged to
advertise shorter prefixes, with consequent greater risk of
them being overridden, there really needs to be some way that the
controllers of routing policy (who filters what) can determine
that they're doing the "right thing" when they refuse to
accept a longer prefix, but still not longer than any prefix
length limits, and route according to a shorter one instead.  Or
that they should allow the longer prefix to override a shorter
one.

History of routes may allow some of that to be automated, but
that is never going to be sufficient.

What's needed is a list of just where routes should be accepted
from - which equates pretty much with address owneship, though
says nothing about the permanence of that ownership.

kre