FritzAdalis@infosec...
replied 24 Jul 2025 12:19 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/yL4DCK6T2GPRq76LJp
benjojo
replied 24 Jul 2025 12:21 +0000
in reply to: https://infosec.exchange/users/FritzAdalis/statuses/114908172234148578
@FritzAdalis "It depends", QUIC's (simplified) origin story is that a bunch of large co's were having the performance of their websites held back by TCP and Windows/Linux/Darwin's kernel implementations, so they made their own TCP running over UDP (to force everything into user space, for full control). The idea to have kernel QUIC is funny in this context because it's "final" step of moving everything back to what it was like before
wolf480pl@mstdn.io
replied 24 Jul 2025 13:03 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/x355Q4q8wj4GfdmZhD
benjojo
replied 24 Jul 2025 13:58 +0000
in reply to: https://mstdn.io/users/wolf480pl/statuses/114908346033969698
@wolf480pl @FritzAdalis Sorta, I would say that SCTP is a exceptionally different transport compared to everything else, so it's not what I would compare anything with (and I say this as a reasonably heavy SCTP user for bgp.tools) QUIC is basically a everything feature transport at this point, and given that was the point originally (aka, innovating beyond TCP) it makes sense ! The only problem is what supports what in server/client etc. There is defo a universe where QUIC could/should have just stayed as the Google transport protocol rather than standardisation, as modern day QUIC still has considerable drawbacks, I suspect most people do not see a huge impact in it's implementation in most cases
srtcd424@mas.to
replied 24 Jul 2025 12:31 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/x355Q4q8wj4GfdmZhD
@benjojo
I think working around "protocol ossification" caused by middleboxes not coping with e.g. new TCP options was also a concern?
@FritzAdalis
bencc@morehammer.uk
replied 24 Jul 2025 13:03 +0000
in reply to: https://mas.to/users/srtcd424/statuses/114908221532713592
@srtcd424 @benjojo @FritzAdalis Oooh had a sudden flashback to diagnosing one of those problems I ended up writing about in a past job: https://blogs.kent.ac.uk/unseenit/stalled-scp-and-hanging-tcp-connections/
srtcd424@mas.to
replied 24 Jul 2025 12:30 +0000
in reply to: https://infosec.exchange/users/FritzAdalis/statuses/114908172234148578
@FritzAdalis (Though actually, mapping NICs into user space and doing everything there is now a thing for high speed server stuff - the important thing is to avoid context switches, the direction doesn't matter!)
I think this diagram means in the sense of development pace, innovation, etc
@benjojo
evey@chaos.social
replied 24 Jul 2025 12:07 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/yL4DCK6T2GPRq76LJp
@benjojo I still love the HEP solution for this all, we just make sure we don't drop packets and then TCP works fine 🙃
amigaunicorn@dudes.k..
replied 24 Jul 2025 12:37 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/yL4DCK6T2GPRq76LJp
sandro@c3d2.social
replied 24 Jul 2025 12:48 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/yL4DCK6T2GPRq76LJp
astraleureka@social...
replied 24 Jul 2025 13:09 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/yL4DCK6T2GPRq76LJp