[ppml] IPv6/IPv4 route table size compared

On Thu, 12 Jul 2007, Stephen Sprunk wrote:

> Thus spake "Dean Anderson" <dean at av8.com>
> > "route table fragmented and inflated by IPv4"???  Let see:
> > IPv4 has 32 bits, IPv6 has 128bits. Which is going to inflate
> > the route table most?  Each route is 4 times larger, and we
> > expect more IPv6 routes.
> 
> We do?  We've been consciously designing allocation policies so that the 
> number of IPv6 routes per AS will be significantly lower than with IPv4.

Yeah, well, there are too few IPv6 routes due to nonuse (Doh!)

Fewer average routes per AS, but more AS's.  (wasn't planning to get
lost on aggregation issues, so I renamed the subject)

> If we end up with more IPv6 routes, it's because the artificial limit on the 
> number of ASes has been raised an order of magnitude.

How about 64k times larger? 32 bit AS numbers (vs current 16 bit AS
numbers)  are under development/deployment. (I haven't been following
ASN expansion, so I don't know exactly where the ASN project stands.

> > I think it won't be long until IPv4 takes up a small fraction of
> > router memory:  200,000 IPv6 routes take up more memory than 200,000
> > IPv4 routes.
> 
> Of course.  However, if we have 200k IPv6 routes, one would expect 2M+
> IPv4 routes, and IPv4 will still end up taking more memory.

I don't think that is likely to happen. The rough math of 2M IPv4 routes
requires an average block size of /21.  (anyone know the average block
size and variation for the current table?)

I can't claim any special knowledge, but I think there is some under
estimation of the number of IPv6 routes, and what root factors cause
routes to be added.  I think we will continue to see more organizations
with allocations because for example, most end users really don't like
having to change IP address numbers or phone numbers. I expect end users
will be seeking something similar to phone number portability in there
internet numbers. That prompts end-users to seek their own allocations
and AS numbers.  This line of thought can be further developed for some
interesting conclusions.

I expect both IPv6 and IPv4 will tend towards fewer routes per AS. This
doesn't mean fewer routes. As long as each AS has both IPv6 and IPv4
needs, those numbers will tend to about the same, with V6 being lower
for a long time. Eventually, I expect that IPv6 will start to outnumber
IPv4 as native V6 users begin to find no use for v4.  But that will take
some long time, I think.

I expect IPv4 mapped IPv6 routes will exist for a long, long time.

> > People seem to anticipate that IPv6 will probably see several
> > million routes, while IPv4 might not ever see 400,000.  It will
> > never be _necessary_ to remove IPv4,
> 
> As we get closer to (and past) exhaustion, the number of IPv4 routes
> is going to explode as people get more, smaller blocks instead of
> aggregates, making the routes-per-AS figure even worse than it already
> is.  

Could be. But it isn't exhaustion that drives that route growth (smaller
blocks); Its multihoming. I'd guess that 32bit AS numbers and
multihoming will make the number of routes per AS drop, but will be more
than offset by the increased number of AS's with small blocks.

		--Dean


-- 
Av8 Internet   Prepared to pay a premium for better service?
www.av8.net         faster, more reliable, better service
617 344 9000