> And to Keith - yes, apps need to get a lot smarter (or perhaps a lot dumber)
> about their use of addresses.   That's true now with IPv4 - we've been using
> temporary use addresses ever since we started using dhcp for address
> assignment.   Addresses come with a lease length, after which they might
> be renewed, or might not be.   Applications, and application protocols
> simply need to deal with that.  Currently we're surviving largely by
> blind luck - that is, the applications that most people use need an address
> to be valid for (at least) an order of magnitude less time than typical
> address validity (perhaps commonly 2 orders).   So, applications are just
> ignoring the issue completely.   To be correct, they shouldn't.

I've come to the opposite conclusion - that it's completely unreasonable 
to expect apps (or hosts) to assume the burdens of:

- selecting the "right" address from several alternatives without knowledge 
  of their own location in the network topology and that of all of their
  current and future peers, 
- dealing with multiple addressing realms (limited-scope addresses are 
  another form of this), where the components of an app are not all in 
  the same realm, or
- expecting apps to know enough about network topology to explicitly
  tunnel through (or configure) NATs, and
- continuing operation across address changes without either support in 
  lower layer protocols to do this or a reliable means of propagating
  address change information to everyone who has an old address
  (they don't necessarily have an open connection)

and that to the degree that apps were ever expected to do this, the folks
who expected them to do those things were on drugs.

it's not just that the apps are being expected to make decisions in the
absence of adequate information to make those decisions (as in,
which of the following N addresses is the best one to use?),
it's also unreasonable to impose burdens on every app that are better
handled by the network.  (e.g. the routing system is having scaling 
problems, so we'll let hosts or apps do address selection to avoid 
having multihomed sites defeat provider-based addressing - never mind 
that they are even less prepared to do routing than the routing system.)

We're not "surviving" by blind luck at all - we're "surviving" by 
pretending that the only apps that matter are those which only need
addresses for a short time (e.g. "web and email are typical")
and ignoring the apps that fail when this assumption doesn't hold.

Keith
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to