Re:40 millisecond bug
by Ono-Sendai on Wednesday November 18, 2020 @03:14AM
(#25134284)
Same old bug, same old blog post repost: http://forwardscattering.org/post/3 "Sockets should have a flushHint() API call"
[ Parent
| on HN ]
This can be accomplished with the TCP socket options TCP_CORK on Linux and TCP_NOPUSH on FreeBSD and OpenBSD. When enabled only full TCP frames are sent; upon disabling any partial frame is immediately sent. See tcp(7) on Linux, tcp(4) on FreeBSD, and http://dotat.at/writing/nopush.html clarifying the behavior.
Re:40 millisecond bug
by Ono-Sendai on Thursday November 19, 2020 @12:10AM
(#25146433)
Yeah, TCP_CORK would work but it's not an elegant solution IMO.