Subject: Re: making our tcp/ip a strong-end system
To: None <sommerfeld@orchard.arlington.ma.us>
From: None <Havard.Eidnes@runit.sintef.no>
List: tech-net
Date: 11/18/1998 18:52:18
> > Hm, if I have understood what the "strong host model" is about, I
> > think there is a place for a "strong router model" too.  The
> > corresponding function in a router would be to refuse to forward
> > a packet entering an interface if the router did not have a route
> > for the source address in the packet pointing back out the same
> > interface the packet entered on.
>
> two comments:
>
> 1) I'd call that a "leaf router" (or maybe "branch router") model, as
>    it assumes symmetric routing, which isn't a reasonable assumption
>    in the presence of redundant paths.

Well, I agree to some extent (about what it could be called).
However, I don't think the description above is precise enough to
describe the situations where use of such a facility would be
inappropriate.  My take:

 o A router can have redundant paths and do equal-cost load sharing
   over those paths.  The "strong router model" would still be
   applicable.

 o A router has redundant paths, used primarily for backup, but
   symmetry of traffic patterns is still maintained.  The "strong
   router model" could still be used here.  During a routing
   fluctuation some packets may "unnecessarily" fall on the floor,
   but "so what"?

 o In a "core router" setup, traffic patterns are often asymmetric.
   In such a situation the "strong router model" sketched above
   would be inapporpriate.

> 2) "leaf-router" vs. "strong host" vs.  .. status should be on a
>    per-interface basis.

I definately agree on this, at least on the "strong router" thing (I
haven't thought much about the "strong host" model).

- H=E5vard