mailing list archives
Re: ? re: Tony Bates' CIDR Report
From: "Sean M. Doran" <smd () clock org>
Date: Sat, 5 Jul 1997 16:17:35 -0700
Not that this is really all that NANOG relevant, but if one checks
into the archives on the NANOG and RIPE mailing lists, you will find
that after lots of discussion, I agreed with Daniel Karrenberg and many
others that /19s were eminently reasonable MAXIMUM prefix lengths for
globally-visible NLRI. A minimum aggregation boundary leading to
no prefixes larger than 18 bits is desirable, and any aggregation
that leads to even shorter global prefixes is a big plus for scalability.
Consequently, just to repeat myself, /19s for everyone meeting
registry requirements is probably just fine. Now let's eliminate
globally visible /24s. All of them. ESPECIALLY the historical
stuff in the swamp and toxic waste dump.
BTW, this probably can serve as notice to people that the step
from proxy aggregation of long prefixes to outright lNAT for offenders
is not that far in the future...
Of course, the plus then is that people at the edges can be as sloppy
as they want wrt routing announcements; from the perspective of large
chunks of the topology (peers' and suppliers' networks, for example),
even rather heavily meshed networks could and will appear to be
neatly aggregated into PA address space.
P.S.: Those vendors who are crowing about flow-directed shortcuts
for IPv4 would be wise in investing in other flow-directed
technologies, such as lNAT, intrusive web caching and redirection,
and the like. I am fairly confident that the near future will
bring a popular recognition among large transit providers (that
is, your initial market) of the Internet as a services-based network
with congestion avoidance principally performed at the edges, with
unknown spaghetti (odd IP topologies, address and protocol translations
and gatewaying, etc.) in the middle. All this nonsense about
addressing MUST go away, and an ability to stage migrations to
post-IP transport protocol(s) is also very handy. Developing a
mindset wherein IP itself is not necessarily end to end, IP
addresses may change relative to topology, and may change over
time is a really keen way of accomplishing that.