Biggest thing I'd change: Build in the ability to do long-duration keepalives from the git-go to avoid three-way handshakes when you want to send data later af

Biggest thing I'd change: Build in the ability to do long-duration keepalives from the git-go to avoid three-way handshakes when you want to send data later after establishing the connection. A whole bunch of other changes to reduce latency in general - no one really quite realized how important latency was back when this stuff was designed, but in all fairness, it's almost, "how could they?" Lastly, extend DNS to provide not only encryption and non-repudiation, bu most importantly, more info, or just absorb Project Athena's Kerberos and Hesiod into DNS at first opportunity. This also allows AFS or other global distributed filesystem support, which in turn could have changed database architecture and semantics for the better. Imagine how different the net would be with a scalable global federated name and directory service that could do everything that Yellow Pages did in the 90s. (N.B.: When I was at Chevron, we built our own version of YP that was hierarchical and multi-domain. It worked really well ("really well", as in damn near flawlessly from the Monday morning we turned it on after a coffee-fuelled weekend hacking session by one of the three true geniuses I've ever met!), seamlessly syncing the info that was in Hesiod/Athena, YP, and DNS to make as much of it as made sense available to clients of each. I've never seen anyone else ever do anything like that, even in all my consulting exposure to other big companies' network services architectures.)