Re: int-serv over e.g. ATM

[Steve Deering <[e-mail]>]

> On another related point, if we look at some of the recent
> dialog, it's obvious that people are carrying around
> different assumptions about their networks.  Here's a few
> pictures of the world (some people call these things
> "reference configurations"):
> 
> 1)	ES - Rtr(s) - ES
> 2)	ES - LAN - ES
> 3)	ES - LAN - Rtr(s) - LAN - ES
> 4)	ES - LAN - Pub-ATM - LAN - ES
> 5)	ES - LAN - Pvt-ATM - LAN - ES
> 6)	ES - LAN - Pvt-ATM - Rtr(s) - LAN - ES
> 7)	ES - LAN - Rtr(s) - Pub-ATM - Rtr(s) - LAN - ES
> 8)	ES - Pvt-ATM - Rtr(s) - Pvt-ATM - ES
> 9)	ES - Pvt-ATM - Pub-ATM - Pvt-ATM - ES
> 10)	ES - Pvt-ATM - ES

Mark,

From an IP perspective, the model is:

        ES - link [- Rtr - link]* - ES

where [foo]* means zero or more repetitions of foo, and each link may be
any network technology (e.g., LAN, private or public ATM, X.25, PPP, or
even an IP-over-IP tunnel).  Concatenations of ATM networks without
intervening routers look like one link to IP, just like LANs concatented
by bridges.

(The above model is true for unicast only, of course.  The multicast case
is harder to illustrate, but instead of being a path it is a tree, with
arbitrary branching and arbitrary branch lengths.)

Any number of ATM links (private or public, in any order) may appear in a
path, each separated by any number of non-ATM links, and the ESs may
separated from any ATM links by any number of non-ATM links.

Designing for anything other than this most general model would be
unnecessarily restrictive and exceedingly foolish.

If there happens to be an IP application that requires a QoS reservation
(I personally have my doubts that such applications need exist, but I'll
suppress that for now), it uses RSVP to make the reservation end-to-end;
at each IP node along the path, including the initiating ES, the module
that processes RSVP invokes link-specific mechanisms (e.g., ATM signalling)
to satisfy the reservation across the next link in the path.  The IP
applications have no need to know about, let alone directly invoke, the
link-specific set-up mechanisms.

This is very elementary stuff that shouldn't have to be said, but your
message made me think that it might need saying, much to my surprise.

Steve