svn commit: r232345 - user/andre/tcp_workqueue/sys/netinet

[Ivan Voras]

2012/3/5 Andre Oppermann <andre at freebsd.org>:

> Thing is they don't shoot into their own foot but everybody else's foot
> in a non-obvious and difficult to debug manner.  While it is user-changeable
> in Linux it not through the normal easy /proc path equaling a sysctl on our
> side.

No, judging by the document, in Linux it's a much more convenient
per-route setting:

" the initial congestion window is configured using the intitcwnd
option in the ip route command. "

http://linux.die.net/man/8/ip

And it apparently was there for a long time (at least since 2005).
This whole thread is very interesting to read:

http://kerneltrap.org/mailarchive/linux-netdev/2010/5/6/6276487

... but if you don't have the time to read it, the tl;dr is: Linux had
the option for years, it's more convenient that what you propose,
off the shelf devices also have it and actively use it, and no
apocalypse has happened.

As I see it, this has stopped being a "purity of spec compliance"
issue since at least 2005, and any attempts to stop the floods now are
at least useless and probably harmful as it leads to "FreeBSD is slow".