(ngtrans) Re: 6BONE AUP

[Robert Elz]

    Date:        Fri, 19 Feb 1999 00:22:49 -0800 (PST)
    From:        Pedro Marques <roque@cisco.com>
    Message-ID:  <14029.6072.69881.109989@pedrom-ultra.cisco.com>

  | 2) Express the router's policy in terms of objects off-line and
  | periodicly resolve the mappings and feed the resolved policies to the
  | router.

This is way too much over specified into implementation detail.

Eg: consider a "routing system", consisting of a traditional router, and
a general purpose host in a box connected to the router over one of its
local net internaces (or any other way).

Let the host be the thing that does the resolving of the mappings, and
the feeding to the router.   That seems like it would fit your model, it
might even be yourintent, and certainly seems like it is a way
that could work.

Now let us notice that the particular router in question is in a big box,
and has a big power supply, and that we don't really need quite a lot of
the associated peripherals for the host for this particular purpose, so
let's move the host processor, and net interface inside the router's big
box.   This is just a matter of packaging, so this can't make a difference
to the model.

Now let's avoid the unnecessary net interface by connecting the host that
is in the box anyway to the router's backplane - a bunch of hardware glue
is needed there, but the model doesn't change.

Now let's notice that we have this fairly general purpose host sharing
backplane access with the router, so let's decide we can also use that
as the the "user interface" into the router - sinking telnet connections,
handling SNMP queries, ... getting the info it needs into and out of the
router as it needs over the backplane.   As I see it, the model still
hasn't changed.

Now let's decide that miniaturisation is the way to go, and build the router
and the host onto a single motherboard, and mount it all in a much
smaller box.  The model surely hasn't changed - but now by any common
perception we have the router doing its own resolving of names into
addresses/masks - and nothing that looks at all off-line involved at all.

Which of these mechanisms any particular router vendor decides to implement
is, of course, entirely up to it - we ought not be attempting to
specify the mechanisms by which it is done.  All that is important is
that it actually be done, somehow, by someone, so users are not left
configuring using IP addresses (and for access lists, filter specs, or
whatever, this is almost as important for v4 as for v6).

kre