Moving past TCP in the data center, part 1 [LWN.net]

Moving past TCP in the data center, part 1

Posted Nov 2, 2022 16:21 UTC (Wed) by jccleaver (guest, #127418)
Parent article: Moving past TCP in the data center, part 1

Curious how many ideas they've revisited from the past here. TCP won due to interconnection ubiquity in the '90s while things from AppleTalk (too chatty, but amazing local discovery) to IPX (painful) went by the wayside, especially as layered on top of IP, but there were a lot of baby ideas thrown out with the bathwater.

Assuming we're keeping IP, I'd be happier to see an old transport protocol with latent potential rediscovered than we move from 14->15 standards.

to post comments

Moving past TCP in the data center, part 1

Posted Nov 29, 2022 20:03 UTC (Tue) by marcH (subscriber, #57642) [Link]

> Curious how many ideas they've revisited from the past here.

I remember discussions about replacing TCP 20 years ago. Should be easy to find if you follow the references in publications. It's like nuclear fusion: it's always 10 years away :-)

This all looks like a very good "state of the union" but commodity and "good enough" seems to win every time. As rich as data center companies are, they're still minuscule compared to the rest of the world using TCP. Moreover their lawyers and businessmen are very likely getting in the way of working together with competitors. It's not enough to tell them "look at how successful Linux is", the obvious answer is "So what? Why would help our competitors catch up with us?"

QUICK has been successful only because Google controls _both_ the browser and the servers - no cooperation required.

Moving past TCP in the data center, part 1

Posted Nov 29, 2022 23:43 UTC (Tue) by marcH (subscriber, #57642) [Link]

> (too chatty, but amazing local discovery)

Off-topic but other-machine.local multicast DNS works most of the time. Often enough that I rarely ever typed an IP address in the last 10 years (and of course you really don't want to type an IPv6 address....)