Re: [Autoconf] Subnet definition

Maybe we are hitting the heart of the misunderstanding. I think we
should indeed distinguish prefix and subnet terms, and stick to
something like:

- a prefix is a contiguous range of IP addresses.
- a subnet is a set of IP addresses (that may or may not be a prefix)
that are on the same link. All these addresses are then on-link with
respect to this link.
- a link is what L2 provides, which reaches everything within TTL 1
beyond a given interface.

If we stick to these definitions, we will indeed save ourselves lots of
trouble. Do you agree?
Emmanuel

On Thu, Nov 20, 2008 at 3:39 PM, Thomas Narten <narten@us.ibm.com> wrote:
I agree that we need a clean definition of subnet. The term is
overloaded and different people have different meanings associated
with it. That has contributed to confusion in the discussions...

Note also that I think Erik's observation about "on-link" being a key
issue to also be clear about is correct.

"Charles E. Perkins" <charles.perkins@earthlink.net> writes:

> A "subnet" is a contiguous range of IP addresses that
> admits a routing prefix.

I'd change that as follows (to highlight the properties that are
important to us):

A "subnet" is the range of IP addresses that are covered by a
prefix. All that a subnet defines is which addresses are considered to
be on the subnet and which are not.

In many cases (in IPv4) there is a one-to-one mapping between a subnet
and a link. I.e., on an Ethernet, you assign a subnet to the ethernet
link.  All the addresses covered by the prefix are assumed to reside
on that Ethernet, and are neighbors of each other. Using IPv6
terminology, we would say "link" rather than subnet and we would say
all the addresses covered by the subnet are "on link". (But note that
in IPv6, you can have links with addresses assigned to it, but for
which there is no prefix that is considered "on link".)

But if one talks to routing folk, a subnet is an abstraction that is
not tied to a physical link. For example, IBM has been assigned net
9.0.0.0. So, to BGP, 9.0.0.0/8 is a "subnet", even though the actual
network is a huge cloud of thousands of individual networks.

And, within IBM, talking about 9.0.0.0/8 as a "subnet" makes little
sense, because we see the individual links that have been subnetted
and given individual subnet numbers. We talk about 9.2.27.0/24 and the
like.

Thus, when talking about a subnet, we also have to consider the
on-link properties (if any), as differing assumptions on that point
can lead to very different ideas about how things do (or don't) work.

Back to the simple Ethernet, all the addresses covered by the prefix
are typically considered on-link.

When talking about MANETs, and talking about subnets (or prefixes) we
also have to talk about the on-link properties (or lack thereof).

For example, going back to chart 37 of ThomasC's presentation, it may
well make sense to consider the entire cloud of routers to be part of
one subnet, but it does not make sense (I assume) to consider that
same subnet to be "on-link" anywhere within that cloud. Within the
cloud, the subnet is further subdivided into smaller subnets. Each of
those individual (smaller) subnets may well have on-link properties
for small portions of the overall cloud.

Thomas
_______________________________________________
Autoconf mailing list
Autoconf@ietf.org
https://www.ietf.org/mailman/listinfo/autoconf
_______________________________________________
Autoconf mailing list
Autoconf@ietf.org
https://www.ietf.org/mailman/listinfo/autoconf