From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Mon Jun 21 2004 - 22:09:48 GMT-3
At 8:29 PM -0400 6/21/04, Scott Morris wrote:
>
>
>Let's move on to other important things like why IPv6 should be dropped from
>the face of the Earth and never be used on a CCIE exam, let alone a real
>network
Let me say that I support the use of IPv6, and indeed take some of
the blame; I was in the plenary meeting at the Toronto IETF that made
the final decision on IPng (among the several candidates). That being
said, I see a great many proposals for using IPv6 to be made for very
bad reasons. There are also some unsolved problems with IPv6
implementation, especially in multihoming, but also in architectural
indecisiveness in separating -- or not separating -- the functions of
locator and identifier.
There are still a lot of people that believe the reason for going to
the 128-bit address space was to have enough room to give a static
address to every insect. Expanding the static address space, in fact,
was one of the non-goals of IPv6 design.
IPv6 consciously, deliberately wastes space in the address field.
This was done, in large part, to avoid the gyrations one goes through
in figuring out how much of a prefix can be summarized. Much of the
IPv6 address space is as-yet unallocated, but such things as the
unicast public address space is designed to be principally
fixed-field. The format ID bits at the start of the address give the
format of the rest, but in a way far more flexible than the four bits
used in IPv4 classful addressing.
Why have fixed-length TLAs and NLAs (and yes, there are nuances of
splitting the NLA)? The simple answer is to make it simple to change
providers. People are generally aware of there being two mechanisms
to get the low-order part of a station's addres: stateful DHCPv6 and
stateless autoconfiguration. In stateless autoconfiguration, end
stations learn the high-order part of their address from the local
router.
What is more of a mystery is how the _routers_ learn the high-order
address. While they can, of course, be statically configured, the
IPv6 suite includes a Router Renumbering Protocol by which a router
can dynamically learn the TLA/NLA. Typically, SLAs will be partially
manually configured (much like OSPF or ISIS areas).
In an ideal scenario, let's say you change ISPs, and you are using
provider-assigned address space that is supposed to change. In IPv4,
this can be quite painful, depending on how well you have prepared
your hosts -- but there are really no good and general mechanisms to
renumber your routers and figure out your new external prefix.
Believe me, I tried -- I wrote
http://www.isi.edu/in-notes/rfc2072.txt , the "IPv4 Router
Renumbering Guide".
But in the IPv6 world, your external router activates a connection to
an ISP router, and the enterprise router plaintively asks, in IPv6
Router Renumbering Protocol, "who am I?" The provider router,
assuming it speaks this protocol, will respond, securely, with the
new high-order bits of the address (basically TLA and NLA for
enterprises). Your external edge routers now propagate this new
prefix into hierarchically lower routers, which pick up SLA bits as
the information propagates downward. Eventually, everything above the
end station identifier is floating around edge LANs in Router
Advertisement packets (ICMPv6).
An end station can then concatenate this prefix to its MAC address or
other locally significant identifier, and, BANG -- it's addressed.
Good practice would have it register the address with Dynamic DNS.
Meanwhile, if you have hosts that don't need direct external
connectivity, they are happily going along in the site-local space
(i.e., IPv6's equivalent to RFC 1918). It may be that the only things
that really need to get new prefixes are the outside addresses of
your firewalls and external router.
Depending on the specific design, you may indeed be able to avoid
overloaded NAT and go one-to-one with provider addresses. This isn't
such a strain on your provider if they can just aggregate the SLA and
below in their customer edge routers, so they only need to advertise
TLA and NLA, or, in large providers, TLA alone, to the rest of the
world.
Do note that we are still experimenting with how to grow the
Internet. While the CIDR assumption has been provider-based
aggregation, IPv6 architecture consciously allows for assigning TLA's
and/or NLA's to geographical exchange points, thus allowing for
geographical as well as provider-based aggregation.
There are other reasons for IPv6, but I wanted to touch on some of
the ones I find are least well understood.
This archive was generated by hypermail 2.1.4 : Sat Jul 03 2004 - 19:40:46 GMT-3