jaeyson · October 7, 2025 11:09 · Oct 7, 2025
diff --git a/99-sysctl.conf b/99-sysctl.conf
@@ -0,0 +1,107 @@
+# etc/sysctl.d/99-sysctl.conf
+# This control is used to define how aggressive the kernel will swap memory pages.
+# We will lower the number to decrease the amount of swap.
+vm.swappiness = 10
+
+# This variable controls the tendency of the kernel to reclaim the memory which is used for caching of directory and inode objects.
+vm.vfs_cache_pressure = 200
+
+# This value in 100'ths of a second define when dirty data is old enough to e eligible for writeout by the kernel flush threads.
+vm.dirty_expire_centisecs = 500
+
+# This value in 100'ths of a second expresses the interval between the kernel flusher wake up period to write old data out to disk.
+vm.dirty_writeback_centisecs = 250
+
+# This is a percentage of the absolute maximum amount of system memory that can be filled with dirty pages before everything must get committed to disk.
+vm.dirty_ratio = 10
+
+# This is a percentage of system memory that can be filled with “dirty” pages — memory pages that still need to be written to disk, before flushed in a background process.
+vm.dirty_background_ratio = 5
+
+# Let's NOT overcommit memory when using VMs (used to be for GitLab Redis)
+#vm.overcommit_memory = 1
+vm.overcommit_memory = 0
+
+# Enable huge memory pages (improving MariaDB and PostgreSQL performance)
+vm.nr_hugepages = 9000
+
+# Network tuning
+
+# Optionally, Disable IPv6
+#net.ipv6.conf.all.disable_ipv6 = 1
+#net.ipv6.conf.default.disable_ipv6 = 1
+#net.ipv6.conf.lo.disable_ipv6 = 1
+
+# This value influences the timeout of a locally closed TCP connection.
+net.ipv4.tcp_orphan_retries = 1
+
+# The length of time an orphaned (no longer referenced by any application) connection will remain
+net.ipv4.tcp_fin_timeout = 20
+
+# Enable memory auto tuning
+net.ipv4.tcp_moderate_rcvbuf = 1
+
+# Turn off timestamp generation, reducing TCP performance spikes
+net.ipv4.tcp_timestamps = 0
+
+# Contains three values that represent the minimum, default and maximum size of the TCP socket receive buffer.
+# Increase default and max. values for both read & write buffers for 10 Gigabit adapters.
+
+# Buffer up to 64Mb for 10 GbE
+net.core.rmem_max = 67108864
+net.core.wmem_max = 67108864
+net.core.rmem_default = 4194304
+net.core.wmem_default = 1048576
+# Increase autotuning TCP limit to 32MB
+net.ipv4.tcp_rmem = 4096 87380 33554432
+net.ipv4.tcp_wmem = 4096 65536 33554432
+
+# Application buffer is 1/..th of the total buffer space specified in the tcp_rmem variable. 
+net.ipv4.tcp_adv_win_scale = -2
+
+# Minimal size of receive buffer used by UDP sockets in moderation.
+net.ipv4.udp_rmem_min = 8192
+# Minimal size of send buffer used by UDP sockets in moderation.
+net.ipv4.udp_wmem_min = 8192
+
+# Maximum ancillary buffer size allowed per socket. 
+net.core.optmem_max = 25165824
+
+# recommended default congestion control is htcp
+#net.ipv4.tcp_congestion_control = htcp
+# Try BBR
+net.ipv4.tcp_congestion_control = bbr
+
+# recommended default for hosts with jumbo frames enabled
+# will prevent block hole, and have no impact on other TCP connections.
+# We will NOT use jumbo frames for now!
+#net.ipv4.tcp_mtu_probing=1
+
+# The default queuing discipline to use for network devices.
+# Instead of fq_codel try just fq
+net.core.default_qdisc = fq
+
+# The maximum number of packets queued in received state
+net.core.netdev_max_backlog = 30000
+
+# Timeout closing of TCP connections after 30 seconds.
+net.ipv4.tcp_fin_timeout = 30
+
+# Avoid falling back to slow start after a connection goes idle.
+net.ipv4.tcp_slow_start_after_idle = 0
+
+# Enable Forward Acknowledgment, which operates with Selective Acknowledgment (SACK) to reduce congestion.
+net.ipv4.tcp_fack = 1
+
+# Support windows larger than 64KB.
+net.ipv4.tcp_window_scaling = 1
+
+# Prevent against common 'SYN flood attack'
+net.ipv4.tcp_syncookies = 1
+
+# Number of times SYNACKs for a passive TCP connection attempt will be retransmitted.
+net.ipv4.tcp_synack_retries = 2
+
+# Maximal number of remembered connection requests, which have not received an acknowledgment from connecting client.
+net.ipv4.tcp_max_syn_backlog = 4096
+
diff --git a/README.md b/README.md
@@ -0,0 +1,9 @@
+My additional kernel setting.
+
+Enables huge pages for better MariaDB and PostgreSQL performance. Support 10Gbit Ethernet adapter. Other various kernel configs like swappiness and dirty ratio.
+
+**Important:** Be sure the values are **not** overridden in your `/etc/sysctl.conf`, because first the `sysctl.d` directory containing configuration files (with values) are read and finally your `/etc/sysctl.conf` file is read. This which could override set values to lower numbers again.
+
+If you enable large pages in Linux, try to disable `transparent_hugepage`, because transparent huge pages are allocated dynamically during runtime, and they are swappable (which is not good for database systems).
+
+[More info about huge pages, transparent hugepages and file system](https://www.enterprisedb.com/blog/tuning-debian-ubuntu-postgresql)
No results found