>> most applications do not use SRV records. > Is it not true that those that use URLs and "connect by name" don't need > to know whether a remote-endpoint address + port is obtained from an A, > an AAAA, or an SRV record?
that's completely false. each application is defined to work in a certain way, and most are not defined to use SRV records. furthermore the SRV RFC explicitly forbids using SRV records for apps that aren't defined to use them. also, each implementation of an application is different, so hacks that work with one implementation might not work with another. >> if they did, they would be >> less predictable and less reliable than they are now because DNS is so >> often out of sync with reality. > Well, there may be bugs to fix, but avoiding completely to depend on the > DNS would seem strange to me. adding useless complexity and making the behavior of both the network and applications less predictable and less reliable seems strange to me. >> it gets even worse because of >> multi-faced DNS because the addressing domains and the visibility of >> those DNS records do not match up, and DNS RRs leak outside of the >> realms in which they're advertised. >> > This is precisely why using only _global_ addresses simplifies a lot: no > need to depend on double-faced DNSs, which tend indeed to create > problems, but because of their double face. that's right, using global IP addresses simplifies things a lot. but trying to impose global DNS names on top of a space where each realm has its own addressing does not simplify things - it makes them far, far more complex. >> advice to everyone working in the NAT space: stop trying to second guess >> application > Note that this proposal is not really "in the NAT space". > It is part of a model where e2e connections no longer need to be > disturbed by NAT traversals. but it doesn't actually solve that problem. > Now, what is you opinion on "get host by name"? I'll believe in applications using only host names when (a) there is a global, secure, scalable, reliable, and fast name resolution system that is everywhere tightly coupled with the underlying addressing and routing system, and (b) when hosts' network stacks use host names rather than addresses as connection/association endpoint identifiers. Until then, host names are just for user convenience. And note that in the present world, a host name means slightly different things from one app to another. _______________________________________________ nat66 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nat66
