[Madlug] A piece of history

[Dale W. Carder]

On Jan 29, 2008, at 3:06 PM, Timm Murray wrote:
> On Tuesday 29 January 2008 02:49:20 pm Marcin Antkiewicz wrote:
> <>
>> Somehow, it makes me thing that, if IPv6 didn't make it in 15 years,
>> while the Internet infrastructure only grew bigger, it should be  
>> let die
>> (just as the OSI stack).
>
> The problem is that we absolutely need it for the Internet to continue
> working.  As the usage of the IPv4 addresses grows, so does the  
> strain on
> core routers.  NAT isn't a long term solution to this problem, and  
> comes with
> its own limitations.

There are sort of 2 separate issues: 1) running out of IP addresses,
and 2) size of core router tables.

#2 isn't really getting addressed in v6 yet either.  Today the v4
DFZ[*] table size is ~245k routes.  The main reasons for this bloat
are deaggrigation via multihoming, traffic engineering, and some
amount of misconfiguration.

The reason this problem is important is for high-speed routers to
forward packets, they need to be able to look up the packet
destinations very, very fast utilizing exotic hardware such as
TCAM[2] or massively parallel processing Tree-Bitmap algorithm[3].

The v6 design sort of implies that the only routes in the v6 DFZ
should be ISP's, and you must get space delegated from your ISP's
IPv6 /32.  If you have multiple ISP's, it was thought you could
use multiple addresses per host.

With the amount of multihoming in particular that has occurred in
the last 10 years, provider-independent IP space will probably
end up a reality in ipv6 for legitimate business reasons.  The
above problem will not get solved with ipv6 as it was assumed
it would maybe a decade ago.

So, with IP address shortage "solved" by ipv6, there is much more
work to be done with the size of the routing tables.

Dale

[1] Default Free Zone, a term for "core" internet routers that
do not have a default route because they have full reachability
information to every AS (autonomous system, aka: IP address
holders like ISP's and companies with their own unique space)
via the BGP protocol.

[2] http://en.wikipedia.org/wiki/Content-addressable_memory#Ternary_CAMs
While cool and very fast they are extremely power hungry,
dissipate a lot of heat, and more importantly for routers
going forward: too hard to reprogram as routes change on
the fly.

[3] One of the biggest routers in the world has an ASIC on
each linecard with (188) 32-bit 250MHz CPUs with a pile of RAM
running this:
http://www.cs.ucsd.edu/~varghese/PAPERS/willpaper.pdf