On Tue, 28 Oct 2003, Matthew Kaufman wrote:
End-to-end requires that people writing the software at the end learn about buffer overruns (and other data-driven access violations) or program using tools that prevent such things. It is otherwise an excellent idea.
A lack of end-to-end just obscures the problem, it does not remove it. Host based firewalling or tcpwrapper style approaches address this point just as well, if not much better. At least if systems shipped with them defaulting to DENY.
Unfortunately, the day that someone decided their poorly-designed machine and operating system would be safer sitting behind a "firewall" pretty much marked the end of universal end-to-end connectivity, and I don't see it coming back for a long long time. Probably not on this Internet. IPv6 or not.
Unfortunate exceptions to the correct design methodology are not an acceptable reason to ignore the correct solution. Most NAT workaround methods currently used by applications fail horribly when both endpoints are behind a NAT, so we are only beginning to feel the initial impact of our reality of slightly broken end-to-end. Imagine an internet that never had end-to-end as a goal... where 'circuits' had to be manually provisioned across multiple carriers networks... where address translation happens at every intra-agency link. Oh wait, we call that the public switched telephone network... and I'm sure we're all already well aware of the amount of innovation that infrastructure affords us, and it's highly economic pricing model as well! I think it's amusing that I see the largest arguments against end-to-end coming from people who ran the networks that the end-to-end internet made largely obsolete.
Combine that with ISP pricing models (helped by registry policy) that encourage <=1 IP address per household, and the subsequent boom in NAT boxes, and the fate is probably sealed.
ISPs simply respond to demand. We're all market whores on this list. Where there is a competitive advantage to offer multiple IPs per customer the ISPs will provide it: we already see this in highly competitive markets. Providing many IPs per household requires no major change to infrastructure, it's simply a policy decision.