[ppml] Policy Proposal 2003-15: IPv4 Allocation Policy for the

[Geoff Huston]

At 02:43 PM 23/09/2003 -0700, bmanning at karoshi.com wrote:
> >
> > >> clearly there is a big promo campaign in south africa to make
> > >> this proposal seem widely supported.  i suggest that input from
> > >> other folk would be interesting.
> > > it was definately widely supported by south africans at the meeting.
> > > unfortunately there weren't many faces from isp's from neighboring
> > > countries present. hopefully they'll make their voices heard, as i
> > > expect there to be many worse off than those of us in .za
> >
> > hint: this affects the GLOBAL routing tables
>             ^ may
> >
> > randy
>
>         since address delegations do not always routing
>         entries make. ...


If one places an interpretation on the comment that this may
have a potentially _negative_ effect on the routing system, then......

Even so, this may or may not have any material outcome
in terms of routing scaleability and stability in any case.

Its not the size of the routing table that has been the problem
for many years - memory is indeed cheap

Nor the size of the forwarding tables - while various forms
of TCAM is not cheap, it appears to be sufficiently
cheap.

It has been argued its the compute load associated with the protocol
update volume, although that has not been a problem for some years

So it has then been suggested that it was the observation that the
growth in the table was faster than Moore's law (doubling every 18 months)
and that the update volume was directly proportional to the size of the table
and this growth _could_ cause the load to exceed silicon capabilities
at some time in the future.

The observation is that growth trajectories have altered substantially 
since 2000
and this form of Internet death by scaling appears to be once more a
very remote possibility.

So when I read this hint I'm inclined to offer the comment that it is at 
present
a  very minor consideration, and that this particular initiative would be 
absolutely
invisible compared with the more prevalent practice of advertising covering
aggregates and then inserting relatively unstable /24s into the global 
routing mesh.
This latter practice is the noisy end of routing that forms the basis of 
our mutual
concerns over scaleability of the entire inter-domain routing system.

Of course the author may well have deliberately avoided any form of
positive or negative connotation in the hint, and was neutrally pointing out
that such a policy would generate additional routing table entries. In which
case, yes, the close correlation between assignment policies and resultant 
routing
advertisements has been noted in the past, and there is no reason to think
this would be any different.

   Geoff