Re: Network layer reqt? [was Re: Transport level multihoming]

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Brian;

> My point is that reliable apps today are built on the basis that TCP
> sessions can break.

Sometimes, they should better adjust TCP parameters and, then,
is a poor and unnecessarily complex implementation.

But, I'm not talking only about reliable apps.

> Furthermore, it is application-dependent machinery
> which would be very complex to build into a transport system.

In such a special case, do something special.

That's my point of end to end multihoming.

> Restoring 
> network level connectivity after an outage is necessary and sufficient.

It is often very complex to build into an application system.

> > Note also that you are assuming large, thus, slow to converge,
> > routing table.
> 
> Unfortunately that is the only safe assumption, until we find a multihoming
> solution that doesn't punch holes.

Considering that there virtualy is NO install base of IPv6, we
can assume anything.

Or, if you say there IS real install base, don't make any
assumption and just count the number of routing table entries
there.

						Masataka Ohta