The computer scientist is optimistic about the future of artificial intelligence. EL PAÍS spoke with him in Washington, D.C., on the 50th anniversary of the pioneering TCP/IP protocols that he co-invented
The TCP landscape is now a hot mess of a multitude of assistance protocols. Maybe we’ll better start from scratch, with modern usecases in in mind and while learning from past mistakes.
This is practically impossible. We have proverbially locked the keys in the room. IP/UDP/TCP are here to stay due to the prevalence of buggy/nonstandard middleboxes (hardware firewalls, ASIC switches, NAT routers) in the Internet.
We have new protocols, new theories for networking (look at NDN, in which is physically incapable of being censored or ddosed). However, anything that doesn’t conform to existing IP/TCP/UDP will get dropped by these so-called “middleboxes”. Even things that DO conform to IP/TCP/UDP will sometimes get dropped by these middleboxes (e.g. new TCP extensions, QUIC, etc). We cannot build an Internet replacement without almost fully scrapping every piece of networking equipment deployed since the 90s.
Middleboxes were supposed to be a temporary solution until we could transition to a new protocol like IPv6. Companies went for the cheap solution and violated the end-to-end principle of networking instead. Now we’re paying the price and stuck with it.
The TCP landscape is now a hot mess of a multitude of assistance protocols. Maybe we’ll better start from scratch, with modern usecases in in mind and while learning from past mistakes.
This is practically impossible. We have proverbially locked the keys in the room. IP/UDP/TCP are here to stay due to the prevalence of buggy/nonstandard middleboxes (hardware firewalls, ASIC switches, NAT routers) in the Internet.
We have new protocols, new theories for networking (look at NDN, in which is physically incapable of being censored or ddosed). However, anything that doesn’t conform to existing IP/TCP/UDP will get dropped by these so-called “middleboxes”. Even things that DO conform to IP/TCP/UDP will sometimes get dropped by these middleboxes (e.g. new TCP extensions, QUIC, etc). We cannot build an Internet replacement without almost fully scrapping every piece of networking equipment deployed since the 90s.
Middleboxes were supposed to be a temporary solution until we could transition to a new protocol like IPv6. Companies went for the cheap solution and violated the end-to-end principle of networking instead. Now we’re paying the price and stuck with it.
I mean, yeah, but most protocols (like DNS, CORS) are a patch on changed usecases or missed consideration.
I’m sure we can vibe something in a few minutes, I’ll solve so many problems when we are forced to use smoke signals for the latest memes.