Everything can always be improved. But there are some things that would make life for a lot of people easier. For one thing, operating systems' tcp/ip stacks.

Everything can always be improved. But there are some things that would make life for a lot of people easier. For one thing, operating systems' tcp/ip stacks. They should have come with TLS as a layer that any app could use just by making the same syscall they use to open a socket. For another, service discovery should not be based on hardcoded numbers, but querying for a given service on a given node and getting routed to the right host and port, so that protocols don't care what ports they use. Domain registration and validation of records should be a simple public key system where a domain owner has a private key, and can sign sub-keys for any host or subdomain under the domain. When you buy the domain you register your key, and all a registrar does is tell other people what your public key is. To prove you are supposed to control a DNS record, you just sign a message with your key; no more fucking about with what the domain owners email is, or who controls at this minute the IP space that a DNS record at this minute is pointing to. This solves a handful of different problems, from registrars to CAs to nameservers and more. The security of the whole web shouldn't depend on the security of 350+ organizations that can all independently decide to issue a secure cert for your domain. The public key thing above would be a step in the right direction. BGP is still ridiculous, but I won't pretend to know it well enough to propose solutions. Give IPv4 two more octets. It's stupid but sometimes stupid works better than smart. Give the HTTP protocol the ability to have integrity without privacy (basically, signed checksums on plaintext content). This way we can solve most of the deficiencies of HTTPS (no caching, for one) but we don't get MitM attacks for boring content like a JavaScript library CDN. And I would make it easier to roll out new internet protocols so we don't have to force ourselves to only use the popular ones. No immediate suggestion here, other than (just like port number limitations) it's stupid that we can't roll out replacements for TCP or UDP. And I would add an extension that encapsulates protocol-specific metadata along each hop of a network. Right now if a network connection has issues, you don't actually know where along the path the issue is, because no intermediate information is recorded at all except the TTL. Record the actions taken at each route and pass it along both ways. Apps can then actively work around various issues, like "the load balancer can't reach the target host" versus "the target host threw an error" versus "the security group of the load balancer didn't allow your connection" versus "level3 is rejecting all traffic to Amazon for some reason". If we just recorded when and where and what happened at each hop, most of these questions would have immediate answers.