[ppml] IPv4 "Up For Grabs" proposal

[James Jun]

[ snip ]

> > > The hitch is: until IPv4 goes away, you're not talking about 100,000
> > > routes. You're talking about 100,000 IPv6 routes PLUS 220,000 IPv4
> > > routes. So it gets worse before it gets better and until IPv4 goes
> > > away, it doesn't get better.
> >
> > It's not that of a big problem than some people are speculating
> actually.
> 
> I'll stipulate that the impending collapse of the IPv4 DFZ has been
> greatly exaggerated if you'll stipulate that IPv6 BGP will converge
> faster when the same routers don't have to converge IPv4 BGP in
> parallel. There will be real incentives not just to move towards IPv6
> but to move away from IPv4.

But currently, the same routers I run today that are participating in both
IPv4 and IPv6 world are not suffering at all -- (and yes it could
theoretically be different when routing table size of IPv6 suddenly bloats
beyond proportion).

May be it's different scenario for you, but for me, I'm spending more money
upgrading my gears to meet growing customer base and their bandwidth demand,
than because of router resource issues created by v4/v6 routing tables.

Either way, it's too early to make any stipulations regarding this in my
opinion as none of us have seen how large_IPv4 + large_IPv6 dual-stacked
routing table would look like in the future.  The fact of the matter is,
major router vendors are making routers with enough control plane horsepower
nowadays[1] that in the future, if things come as predicted, no one will pay
a lot of attention to routing table size anymore at least probably from
carrier positions.

[1] i.e. one good example is the recent RSP720 supervisor to replace SUP720
for Cisco 7600.  MSFC speed is more than doubled.


And then there are elements driving the networking hardware industry on
matters that don't have a lot to do with BGP routing convergence.  Vendors
are continuously competing each other trying to build platforms with new
integrated features, triple-play, converged networking, self-defending
network(tm), CALEA On Demand(tm), 802.3ah Ethernet OAM, virtual routers, etc
yadda yadda, that all require continued improvement of their control plane
hardware and software architectures (see IOS XR and Cat65k modular IOS
code).  Enhanced hardware and software capacities required for
implementation of these new features alone could make routing protocol
convergence issue a non-issue.

A long time ago, people speculated that the Internet will come to grinding
halt if routing table size bloats in excess of 200k, and as ridiculous as it
may sound, some have even questioned whether the trie data structures for
RIB will be able to scale to hold let alone scalability of BGP convergence.
But it seems history has taught us that router companies are in business to
make money by designing and building routers.  And they seem to be aware of
scalability issues in their own products looking ahead in the future --
especially when their customers vote with their money by looking at their
competitors.


But are we going to see technological advantages (i.e. clean routing
table??) in IPv6 that would make IPv4 a liability for an organization to
keep on running?  I'm betting no.  IPv6 is no different than IPv4 in my book
other than increased address space.  Deaggregation, routing table bloat,
security issues, etc, etc we see today will be business as usual in IPv6
land too.  It's just the way it is.

In closing, at the end of the day, IPv6 will most likely prevail, probably
not because of routing table/router resource issues and other problems in
current IPv4 internet, but more likely because we are simply running out of
addresses in IPv4.  Does that mean we should simply shut down IPv4? (as
someone on this mailing list indicates that we should, but I'm not talking
about you ;>) No.  We should rather work on improving mechanisms that permit
easier transition.


James