mailing list archives
Re: number of unaggregated class C's in swamp?
From: Yakov Rekhter <yakov () cisco com>
Date: Mon, 02 Oct 95 15:18:01 PDT
......... Yakov Rekhter is rumored to have said:
] So, perhaps we should just look at the total amount of IP address space
] advertised by a provider in its routing advertisements, then divide
] this amount by the number of routes the provider advertises, and
] see whether the resulting number meets the goal.
But what is the goal?
One goal is to come up with a metric to measure how efficient the
routing system works (especially wrt to aggregation), and how well
individual providers manage to aggregate. This would allow us to look
at parts of the system (e.g. individual providers), and see which of
these parts would need improvements. Also if we have such a metric, we
could look at how various mechanisms/incentives could influence the metric.
My capitalist nature says that the amount of address space one has should
not be an issue. I'm not terribly sure on how that enters into the metric.
I'd be in favor of something that directly associated 'goodness' or 'cost'
with the amount of ip nodes one could route, or the ratio of routes to nodes.
Ideally, as you suggested, it would be really nice to have a metric
that would tell us how efficiently the routing *and* addressing system
works wrt to providing routes to actual hosts, rather than to blocks of
addresses (after all, the purpose of the routing system is to provide
connectivity to hosts, not just to host-less addresses). Moreover, if
we would have such a metric, we might be able to come up with some
mechanisms/incentives that would truly promote scalable Internet, so that
such mechanisms/incentives would both (a) drive towards more efficient
address space utilization (thus imposing a back pressure on consumption
of one finite resource - IP address space), and (b) drive towards more
routing aggregation (thus imposing a back pressure on consumption of
another finite resource - forwarding tables).
In practice, getting this metric requires some way of knowing the
number of hosts per prefix. We don't have any technology to do this, so
we rely on a "simplifying assumption" - we assumed that the amount of
address space that an Internet Registry allocates to a site on average
reflects the number of hosts within the site. I am well aware of all
the traps associated with this "simplifying assumption", but at the
moment that is all what we have.
So, to sum things up, if we measure how efficiently we route wrt to
address space (which we can do today), and if we'll augment this data
with the "simplifying assumption", we can at least get some *rough*
approximation of the "ideal" metric.