The IP over ATM Mailing List Archive by date

[schulter@zk3.dec.com]

On Wed, 17 Jan 96 15:07:00 PST Ran Atkinson wrote:

> I believe that the above clarifies much of the disagreement between me
> (and some others) and Peter/Grenville.
>
> I do not believe that there has ever been consensus that traditional ND
> should/must be used over any kind of link that lacks a native broadcast
> technique (e.g. MARS is not a native technique to ATM).  Things that most
> folks think ND works well over include Ethernet, FDDI, Token Ring, and many
> types of radio links.  Things that many folks (clearly not all, Peter and
> Grenville being examples in the set that disagree with me here :-) believe
> should NOT use the traditional ND include NBMA links and other media
> lacking any native broadcasting techniques.

This is certainly a point which should be discussed and consensus reached,
both in the IP over ATM community, and the IPv6 community.  I include
the IPv6 community in this since I think that saying that ND is not to
be used over NBMA media will have some affects on the network layer. I
have added the IP over ATM mailing list to my reply since folks there
are probably interested in this discussion.

My current view of the IPv6 over ATM world is that, unlike IPv4, IPv6
has a built-in address resolution protocol in ND.  That is, IPv4 choose
to leave address resolution up to the datalink layer (i.e., ARP) and
datalink protocols (well, this wasn't a choice as much as unecessary
for what TCP/IP was originally intended to do; broadcast media were
dealt with later).  This is why the IP over ATM group could define
new protocols like ATMARP and NHRP without requiring any changes to the
network layer.  The network layer simply left it up to the datalink layer
to perform the address resolution approrpiate to the specific media.  On the 
other hand, IPv6 has chosen to include address resolution as part of the base 
network layer protocol suite; that is, ND.  ND is placed at the ICMP layer, 
rather than the datalink layer.  ND uses IPv6 packet headers, IPv6 security,
and manages the IPv6 prefix list and routing tables which are used by
IPv6 to make routing decisions. Further, ND is meant to be media independent, 
so IPv6 expects to use ND for all these functions on all interfaces.
impression that ND can't be "turned off" or made "optional".  Further, ND is 
used for much more than simply address resolution, it is also used for address 
autoconfiguration, router discovery, router redirects, and network 
configuration.  Thus, it seems to me that it is very integral to IPv6 and 
simply can not be replaced with some other protocol on a media-by-media basis.

Given this, my approach to IPv6 over ATM is to design something which
will permit ND to work over ATM with no modification or changes of
any sort to ND or the network layer.  That is, the network layer should
be able to treat all datalinks the same since this is what ND seems to
want to do.  This makes the IPv6 over X media problem exclusively a
datalink layer problem, and this is where I think the problem should be solved.
Of course, solving some of these problems in the datalink means that
some datalinks (like ATM) must deal with a protocol which is inhierently
mismatched to the characteristics of the datalink.  To me, this means
developing a way to adapt ND to the datalink layer rather than changing or 
replacing ND.

Thus, my expectations have always been to try to accomodate ND over ATM
rather than trying to bypass or replace ND.  If ND is bypassed or
replaced this would mean trying to specify a new network layer protocol that 
does address resolution, router discovery, address autoconfiguration, router
redirection, and network configuration (all while trying to keep ND away
from the network), all of which will have to be done if ND is not used on a 
specific media.  Further, if one chooses to say "ND does not apply to
media type X" then some changes to the way IPv6 and ND are currently
defined must be made so that some other protocol can be plugged in where 
ND sits on a per-media basis (or so that ND can be unplugged on a per-media
basis).

So, ND over ATM may look like a mismatch (and it is), but I think it is quite
possible to develop a method of adapting ND to ATM just as some method of
adapting IP multicast to ATM had to be developed (another case in which
a base feature of the network layer protocol was mismatched to the media
characteristics).  Further, I think adapting ND to ATM is a much simpler
proposition that trying to come up with a new protocol that does what ND
does, can be made fairly light weight, and can be made quick and easy to
implement.  And all this can be done without the network layer having to
know anything about it. This is exactly the tact I'm taking in my I-D for 
IPv6 over ATM.

 --- pete

------------------
Peter Schulter					schulter@zk3.dec.com
Digital UNIX Networking				voice (603) 881-2920
Digital Equipment Corp				voice (DTN) 381-2920
ZK3-03/U14					FAX   (603) 881-2257
110 Spit Brook Road				FAX   (DTN) 381-2257
Nashua, NH 03062