4 ms·
>everything got thrown into IPv6 without solving the two problems that actually matters That would arguably be much better suited to a layer 4 solution. SCTP h
by MertsA 6y ago
>everything got thrown into IPv6 without solving the two problems that actually matters
That would arguably be much better suited to a layer 4 solution. SCTP has solved both of those problems and is now 2 decades old. I have in the past harped on what a shame it is that QUIC was built, basically reinventing features of SCTP poorly. SCTP could still be viable over the open internet on IPv6 if it's done before the inevitable protocol ossification happens on IPv6 just like IPv4 when middlebox vendors can't be bothered to support passing traffic other than TCP, UDP, and ICMP.
- xxpor 6y agoSCTP-over-UDP is still totally viable.
- anderspitman 6y agoWhat do you see as the main advantages of SCTP over QUIC?
- MertsA 6y agoFor HTTP? Basically just multihoming. For everything else? Multiplexing streams, multihoming, reliable datagrams, out of order delivery, in order delivery of datagrams, resistance to SYN floods, etc. It's basically a swiss-army knife transport protocol. By far the biggest drawback of SCTP is lack of widespread support. Linux and the BSDs have it implemented in the kernel but Windows and macOS only have third party implementations. It can be implemented in userspace, but now you have to tunnel all of it over UDP instead of just native SCTP and now you need to worry about separate userspace implementations conflicting by trying to use the same UDP port if you want to host it as a server.