Re: initial issues

[Stig Venaas <Stig.Venaas@uninett.no>]

On Thu, Feb 15, 2001 at 11:44:10AM +0100, Erik Nordmark wrote:
> For instance, we have
>  - host multihoming
>  - site multihoming (where site is an administratively
> 	configured well-connected piece of topology)
>  - AS multihoming
> 
> Isn't the main interest really in solving the latter two?

At least the discussion seems to be focused on those. If all hosts in a
site were multihomed, you would in essence have site multihoming, or?

I think the end-to-end multihoming solution should be considered. If you
have SCTP with multiple addresses for endpoints (similar has been suggested
for TCP), you could have the session survive a failure in one of the site's
connections. I've also heard rumors of an SCTP draft for dynamically changing
endpoints but I can't find it now. One could also have more intelligent
applications that are able to save state and continue with a new session.

In my opinion this deserves more attention, it should at least be
considered. I agree we shouldn't discuss solutions yet, but I would like
the scope to include this possibility. There might be good reasons for not
to take this path, but we shouldn't discuss that yet.

Stig