jnc at ginger.lcs.mit.edu
Tue Sep 26 16:17:09 UTC 1995
From: gih at aarnet.edu.au (Geoff Huston)
Given the interaction between the costs of renumbering, vs the cost of
establishing an entry in the routing tables, vs the cost of dynamic
address translation then its not obvious whether pricing policies should
apply to address space allocation, routing table entries or both!
It's a truism of large-scale system design (in both biology as well as
computers :-) that a single mechanism can do two things well at small scales,
but not larger ones.
This says to me that a pricing policy that attempts to meet a number of goals
(e.g. control routing table growth, maximize efficiency of address space
usage) by pricing a single thing is just not going to be as effective,
overall, at those goals, as a number of separate pricing policies on different
resources. (Hmm, maybe that principle above applies to economics as well?
Makes sense, economics and biology have close ties...)
Of course, two separate mechanisms (or indeed, any at all) are a lot more
complexity, administrative overhead, hassle expended in setting policies (look
at DNS name charging :-), etc, etc, so you have to step back and ask, overall,
if the gains in getting closer to those goals (using either one mechanism or
two) are worth what it costs in complexity, etc.
None of which is much help in answering the question "should we price none,
one, or both". What I suspect is that we will price none until we simply
can't go on without charging, which makes sense; the administrative overhead
of charging will be a major headache, so the situation will have to be dire
to make it worth it.
More information about the NANOG