On Tue, Dec 14, 1999 at 12:28:46PM +0100, Harald Tveit Alvestrand wrote:
> At 16:02 13.12.99 -0800, Kent Crispin wrote:
> >On Mon, Dec 13, 1999 at 03:38:44PM -0800, Randy Bush wrote:
> > > > I don't think this is a problem for the routing system.
> > >
> > > indeed, you seem to be pushing a lot of routing decisions to the host. and
> > > i am worried about where in the host's stack those decisions are
> > supposed to
> > > be made.
> >
> >Application layer.
> Not in my 5-line client, it's not.
> Source/destination address choice inside whois? inside finger? NO.
That was a flip remark for entertainment purposes, and it begs the question
of what you mean, precisely, by "application layer" in a non ISO world.
That being said, it would be straightforward to bury something like this in
libraries, so your 5 line application just loads with the multihoming
library code, which would be, in practical terms, an integrated
resolver/socket library that cached performance metrics for IP addresses
associated with a particular domain name, and automatically switched to
different IP addresses when performance became bad. You would want to
have it periodically ping the other addresses, as well.
I'm not proposing from a position of advocacy, you understand, but from a
pure implementation point of view, it seems straightforward, and you could
easily write 5 line clients that used it.
--
Kent Crispin "Do good, and you'll be
[EMAIL PROTECTED] lonesome." -- Mark Twain
