Linux 7.0 Quietly Kills a 38-Year TCP Design Problem - AccECN Is Now On by Default

Discussion
1

Hello Friends

For your consideration

:penguin: :globe_with_meridians:

3 Likes
2

It shows that Linux/BSD developers are the real masters of networking. Others just use it or copy code.

2 Likes
3

The article doesn’t mention BSD - but I agree - it’s a much better “Network Operating System” than many others - hence why many vendors use it to base their system on top of - e.g. NetApp, Juniper, and Apple - to name a few - I’m sure there are many more… and the BSD license is more permissive than the GPL - so vendors don’t necessarily have to “give back”… I seem to remember seeing the string “Berkeley Sockets” in TCP/IP stacks for MS DOS way back in the 1990’s… and I think Microsoft used some of that code to the “TCP/IP Addon for Windows for Workgroups” (before then WFW could only talk IPX or Netbeui).

However - I don’t know enough about this topic to understand if any of the BSDs have implemented AccECN - there’s this article :

but it’s way above my skill level to interpret… it does mention AccECN though…

3 Likes
4

That is where networking started. I bet the article does not even mention Unix either.

Not sure about BSD with AccECN.
Google AI says
" FreeBSD (starting from version 14) supports Accurate ECN (AccECN) as part of its network stack updates. This feature allows for more fine-grained, byte-level feedback about network congestion compared to the traditional single-mark-per-RTT mechanism"
So yes, at least for FreeBSD.
It is in your link too.

3 Likes
5

It says FreeBSD supports it. But the first article says Linux has supported it for a couple years. It just was not the default until the 7.0 kernel.

3 Likes