Linux Speed, Responsiveness, and Latency Reduction with 'sysctl' Settings

article #892, updated 67 days ago

On Manjaro, Debian, Ubuntu, and several others, these items are known help a lot in any application, including desktop, web server, or terminal server. The end of this post has two large compilations of these settings, one for wired (“non-lossy”) networking, one for wireless (“lossy”).

Do not use these on antiX or MX Linux, their developers have done much better right off the ISO. If you want more, recompile the Xanmod kernel for performance.

On the vast majority of Linux distributions, one can just add these changes to /etc/sysctl.conf, and then run sysctl -p to apply them without reboot. However, recent additions to standards have enabled us to place custom settings in our own configuration files, so that we don’t take /etc/sysctl.conf out of distro control.

On recent Debian and Ubuntu, we may best put them in /etc/sysctl.d/99-custom.conf (or replace the word “custom” to your liking), and then run sysctl --system to load both /etc/sysctl.conf and everything under /etc/sysctl.d. The 99 causes it to be run last, i.e., to override any others. But look at what already is there; sometimes there is already a last 99.

On some other recent distros, it’s /etc/sysctl.d/custom.conf (the word “custom” is still arbitrary), and then run systemctl restart systemd-sysctl.

You can check your results with sysctl -A.

The first group is for wired networking performance:

net.ipv4.tcp_window_scaling=1
net.ipv4.tcp_workaround_signed_windows=1
net.ipv4.tcp_sack=1
net.ipv4.tcp_fack=1
net.ipv4.tcp_low_latency=1
net.ipv4.ip_no_pmtu_disc=0
net.ipv4.tcp_mtu_probing=1
net.ipv4.tcp_frto=2
net.ipv4.tcp_frto_response=2
net.ipv4.tcp_congestion_control=illinois

A bit different first group for networking performance, is recommendable for anything involving wireless, i.e., “lossy” networks:

net.ipv4.tcp_window_scaling=1
net.ipv4.tcp_workaround_signed_windows=1
net.ipv4.tcp_sack=1
net.ipv4.tcp_fack=1
net.ipv4.tcp_low_latency=1
net.ipv4.ip_no_pmtu_disc=0
net.ipv4.tcp_mtu_probing=1
net.ipv4.tcp_frto=2
net.ipv4.tcp_frto_response=2
net.ipv4.tcp_congestion_control = hybla
net.ipv4.tcp_allowed_congestion_control = hybla cubic

And then some general networking performance items:

net.core.rmem_default = 31457280
net.core.rmem_max = 12582912
net.core.wmem_default = 31457280
net.core.wmem_max = 12582912
net.core.somaxconn = 4096
net.core.netdev_max_backlog = 65536
net.core.optmem_max = 25165824
net.ipv4.tcp_mem = 8388608 8388608 8388608
net.ipv4.udp_mem = 8388608 8388608 8388608
net.ipv4.tcp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem_min = 16384
net.ipv4.tcp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem_min = 16384
net.ipv4.tcp_max_tw_buckets = 1440000
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_tw_reuse = 1
net.ipv4.route.flush=1

And some for network security enhancement:

net.ipv4.tcp_synack_retries = 2
net.ipv4.ip_local_port_range = 2000 65535
net.ipv4.tcp_rfc1337 = 1
net.ipv4.tcp_fin_timeout = 15
net.ipv4.tcp_keepalive_time = 300
net.ipv4.tcp_keepalive_probes = 5
net.ipv4.tcp_keepalive_intvl = 15

And now a few to keep virtual memory usage under good control:

vm.swappiness=20
vm.dirty_ratio = 60
vm.dirty_background_ratio = 2

And one to increase the maximum number of open and watched files, very helpful indeed for servers, file synchronization of all sorts, and many other functions:

fs.file-max = 2097152
fs.inotify.max_user_watches = 524288

The above was compiled from these two excellent articles:

http://www.networkworld.com/article/2227856/opensource-subnet/best-networking-tweaks-for-linux.html
https://easyengine.io/tutorials/linux/sysctl-conf/

and other sources. Here is the whole set for wired (non-lossy) networking:

net.ipv4.tcp_window_scaling=1
net.ipv4.tcp_workaround_signed_windows=1
net.ipv4.tcp_sack=1
net.ipv4.tcp_fack=1
net.ipv4.tcp_low_latency=1
net.ipv4.ip_no_pmtu_disc=0
net.ipv4.tcp_mtu_probing=1
net.ipv4.tcp_frto=2
net.ipv4.tcp_frto_response=2
net.ipv4.tcp_congestion_control=illinois
pre. net.core.rmem_default = 31457280
net.core.rmem_max = 12582912
net.core.wmem_default = 31457280
net.core.wmem_max = 12582912
net.core.somaxconn = 4096
net.core.netdev_max_backlog = 65536
net.core.optmem_max = 25165824
net.ipv4.tcp_mem = 8388608 8388608 8388608
net.ipv4.udp_mem = 8388608 8388608 8388608
net.ipv4.tcp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem_min = 16384
net.ipv4.tcp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem_min = 16384
net.ipv4.tcp_max_tw_buckets = 1440000
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_tw_reuse = 1
net.ipv4.route.flush=1
vm.swappiness=20
vm.dirty_ratio = 60
vm.dirty_background_ratio = 2
fs.file-max = 2097152
fs.inotify.max_user_watches = 524288

and another full set for wireless / lossy networking:

net.ipv4.tcp_window_scaling=1
net.ipv4.tcp_workaround_signed_windows=1
net.ipv4.tcp_sack=1
net.ipv4.tcp_fack=1
net.ipv4.tcp_low_latency=1
net.ipv4.ip_no_pmtu_disc=0
net.ipv4.tcp_mtu_probing=1
net.ipv4.tcp_frto=2
net.ipv4.tcp_frto_response=2
net.ipv4.tcp_congestion_control = hybla
net.ipv4.tcp_allowed_congestion_control = hybla cubic
pre. net.core.rmem_default = 31457280
net.core.rmem_max = 12582912
net.core.wmem_default = 31457280
net.core.wmem_max = 12582912
net.core.somaxconn = 4096
net.core.netdev_max_backlog = 65536
net.core.optmem_max = 25165824
net.ipv4.tcp_mem = 8388608 8388608 8388608
net.ipv4.udp_mem = 8388608 8388608 8388608
net.ipv4.tcp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem_min = 16384
net.ipv4.tcp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem_min = 16384
net.ipv4.tcp_max_tw_buckets = 1440000
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_tw_reuse = 1
net.ipv4.route.flush=1
vm.swappiness=20
vm.dirty_ratio = 60
vm.dirty_background_ratio = 2
fs.file-max = 2097152
fs.inotify.max_user_watches = 524288

Categories: