Re: [Autoconf] comments on draft-ietf-autoconf-manetarch-07.txt

[Thomas Narten <[e-mail]>]

> Maybe better definition could help.

Well, yes! :-)

> I think it is easy to understand that a MANET Interface is an interface that
> may suffer from certain link characteristics. We have similar kind of
> interface types, like NBMA and P2MP.

No. NBMA is a link type, not an interface type. IPoverATM, for
example, looks to IP just like a traditional IP-friendly link. There
is a layer of software between IP and ATM that emulates the necessary
services that IP needs (like ARP/broadcast). The ATM "interface",
however, is just a normal interface. The IP layer does not know about
the quirks of ATM and doesn't need to worry about the "challenges" in
semantics in order for IP to work.

Another way to view things. Stuff that is "interface-specific" is
normally implemented by a device driver (for the specific link layer),
and not by IP.

> Now we want to define another.
> Term MANET Interface is well known, e.g. it is used in MANET Routing
> Protocol documents, including the OSPF ones.

The term "interface" is also well-known outside of MANET. If you look
at RFC1122, for example, interface is about the interface beween
layers (e.g, between IP and the link layer). The more general term
"interface" (in terms of network software) is the data abstraction
used to access the facilities of a particular network.

> >    MANET Router (MNR)
> >       A MANET router is distinguished by having one or more MANET
> >       interfaces.  A MANET router may also have zero or more non-MANET
> >       interfaces.  A MANET router is responsible for hiding MANETs'
> >       challenging characteristics from nodes that are not MANET-aware.
> >       A MANET router with a single MANET interface is illustrated in
> >       Figure 1.
> > 
> > No. the "interface/link" hides the challenging characteristics from
> > IP. Not the BR itself. This is a critical point (architecturally) to
> > get right.
> > 

> If this is true, we do not have to do anything special.

Well, I'm trying to understand whether this is the case or not.

> > You really (I think) need to define an abstraction that defines an
> > link type of "manet" that handles all the "interesting"
> > characteristics of MANETs (like non-transitive connectivity) but that
> > also presents itself to IP as a single IP subnet that has the normal
> > properties.

> I think there are two MANET implementation models.

> One is hiding MANET form IP, e.g. a sub-IP MANET model. This is more or less
> out-of-scope for IETF.

This could be done in the IETF, and maybe this is  a direction the WG
may want to pursue. But is this even a model the WG is currently
considering?

IMO, this is the _obvious_ model to implement, or rather, if any other
model is needed, the case first needs to be made why the above model
is not good enough (or better).

> The other model is described in this document.

And what model is that? The problem I have currently is I don't
understand the scope of the problem. What is the abstraction. Where is
the MANET-specific stuff implemented, and what is the boundary between
this and what IP sees?

E.g., the MR model suggests that I can have an ethernet interface and
two manet interfaces. Where does the MANET stuff sit (just above the
two MANET interfaces?) What part is visible to IP (and why?) Does the
MANET specific stuff get exported across the Ethernet? Why or why not?

Very basic questions, but the document doesn't make this clear at all.

> Now we have to face the reachability issues. I think the MANET
> Router shall hide this for IP hosts and non-MANET routers.

Hide it from whom? What is the scope of where this information is
supposed to be exported? It's not good enough to say "within the
MR". Where within the MANET router?  What part of the MR code is
"normal, unmodified IP" and what part has to be MANET-aware?

Thomas


_______________________________________________
Autoconf mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/autoconf