Iljitsch van Beijnum wrote:
On 19-sep-2007, at 0:10, Keith Moore wrote:
What bugs me is that I think that the existence of mini-cores (or more
generally, a large number of private interconnections between networks
using ULA prefixes) leads to a world where it becomes important to have
a particular kind of source address to talk to a particular kind of
destination address, and in which applications are expected to choose
the right source address in order to talk to a particular kind of
destination address. So the sources addresses available to a particular
host end up being like a keyring. I don't think that's a good burden to
put on apps, and I don't think that using addresses like authentication
tokens is a good way to go.
I don't think having a large set of source addresses with different
reachability is a good thing, but it's likely unavoidable to have
several.
If applications don't want to worry about addressing issues, the only
solution is that applications don't get to see addresses in the first
place.
If you can find a good way to let hosts and network stacks sort out
those addressing issues, then fewer applications will bother to manage
those addresses themselves. But absent such a way to do that (and
trial-and-error is not a good way, nor is IPv6 "address selection") then
more applications will be forced to manage those addresses in order to
provide decent service to users. Or to put it another way, as long as
an application can provide better service by explicitly managing
addresses than one that doesn't explicitly manage addresses, there will
be incentives for apps to explicitly manage addresses and many apps will
do so.
There's a whole separate issue in that networks trying to do filtering
on the basis of source and/or destination address hides policy from
applications. So the application doesn't know whether it is violating
policy or routing around a failure of the network. And realistically,
the answer is probably different for different applications.
Keith
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf