[arin-ppml] Rationale for /22

[David Williamson]

On Tue, Jul 28, 2009 at 10:03:57AM -0700, Ted Mittelstaedt wrote:
> If you set the minimum size of a table entry to /25 instead of /24 then
> you have now doubled the possible table entries.  It doesn't matter if 
> the entries are gratuitous or not, the way subnet mathematics works, for
> every drop in the bit boundary, from /25 to /26 to /27 and so on, you
> double, quadruple, etc. the total possible route entries.
> 
> With the increase in router ram and router CPU speed over the years, 
> some decrease in the minimum is warranted.  If we didn't have IPv6 
> coming up - which will also double the number of advertisements once
> everyone runs dual-stack - we could support an even further decrease
> in the minimum.  But we can't yet support multiplying the number of
> potential routing slots by 200 on the Internet.

I find your math to be odd.  Yes, there are twice as many /25s possible
as there are /24s, but moving the filter by one bit doesn't immediately
blow up the routing table - very few networks are fully deaggregated.
(And those that are need to be fixed, in my opinion.)  I suspect that
accepting /25s will increase the rate of growth by a bit, but not
radically so.

Also, the ratio of v4 announcements to v6 at steady state better not be
1:1.  We're attempting to be much smarter at handing out v6 space, so
there should be many fewer routes at steady state in v6.  Of course,
they'll consume 4x the memory, so were still a bit screwed, but not as
badly as your math implies.

-David