Limit mss fixup to connection start
Use conntrack flow packet counter to limit mss fixup filter to the very start of connection flows where adjustable syn/syn can possibly appear.
Two initial packets +
sysctl net.ipv4.tcp_syn_retries net.ipv4.tcp_synack_retries net.ipv4.tcp_syn_retries = 6 net.ipv4.tcp_synack_retries = 5
Signed-off-by: Andris PE [email protected]
Other heuristics take twice bytecode like !assured/assured & ct tcp, Rises throughput ~10% in default case both unicast and forward.
Any idea when your fw4 improvements will be accepted into a build with the new build 23.05.5 having no firewall updates? They are still all pending incorporation into master by jow.
Thanks Peter
Message ID: @.***>
He comes here twice a year, if he misses few watch shifts ill make a branch and add readme ;-)
Shouldn't the hardcoded 14 value be calculated dynamically if it relates to net.ipv4.tcp_syn_retries and net.ipv4.tcp_synack_retries ?
Also this looks like very specific, hard to maintain low level optimizations - what's the significance of this change? Are there any throughput tests?
Will adapt in a week
This is about state of internet. Windows limits all retransmits to 5 to https://support.microsoft.com/en-us/topic/how-to-modify-the-tcp-ip-maximum-retransmission-time-out-7ae0982a-4963-fa7e-ee79-ff6d0da73db8 Thus one obtained from linux covers it. If you achieve more than few retransmits under normal operation, mss size is least of your woes.
This can be done even quicker, no point keeping it another yesr