This the multi-page printable view of this section. Click here to print.

Return to the regular view of this page.

Tuned Template

Several pre-defined tuned templates

1 - OLTP

Tuned OLTP模板

Tuned OLTP模板主要针对延迟进行优化,此模板针对的机型是Dell R740 64核/400GB内存,使用PCI-E SSD的节点。您可以根据自己的实际机型进行调整。

# tuned configuration
#==============================================================#
# File      :   tuned.conf
# Mtime     :   2020-06-29
# Desc      :   Tune operatiing system to oltp mode
# Path      :   /etc/tuned/oltp/tuned.conf
# Author    :   Vonng(fengruohang@outlook.com)
# Copyright (C) 2018-2021 Ruohang Feng
#==============================================================#

[main]
summary=Optimize for PostgreSQL OLTP System
include=network-latency

[cpu]
force_latency=1
governor=performance
energy_perf_bias=performance
min_perf_pct=100

[vm]
# disable transparent hugepages
transparent_hugepages=never

[sysctl]
#-------------------------------------------------------------#
#                           KERNEL                            #
#-------------------------------------------------------------#
# disable numa balancing
kernel.numa_balancing=0

# total shmem size in bytes: $(expr $(getconf _PHYS_PAGES) / 2 \* $(getconf PAGE_SIZE))
{% if param_shmall is defined and param_shmall != '' %}
kernel.shmall = {{ param_shmall }}
{% endif %}

# total shmem size in pages:  $(expr $(getconf _PHYS_PAGES) / 2)
{% if param_shmmax is defined and param_shmmax != '' %}
kernel.shmmax = {{ param_shmmax }}
{% endif %}

# total shmem segs 4096 -> 8192
kernel.shmmni=8192

# total msg queue number, set to mem size in MB
kernel.msgmni=32768

# max length of message queue
kernel.msgmnb=65536

# max size of message
kernel.msgmax=65536

kernel.pid_max=131072

# max(Sem in Set)=2048, max(Sem)=max(Sem in Set) x max(SemSet) , max(Sem per Ops)=2048, max(SemSet)=65536
kernel.sem=2048 134217728 2048 65536

# do not sched postgres process in group
kernel.sched_autogroup_enabled = 0

# total time the scheduler will consider a migrated process cache hot and, thus, less likely to be remigrated
# defaut = 0.5ms (500000ns), update to 5ms , depending on your typical query (e.g < 1ms)
kernel.sched_migration_cost_ns=5000000

#-------------------------------------------------------------#
#                             VM                              #
#-------------------------------------------------------------#
# try not using swap
vm.swappiness=0

# disable when most mem are for file cache
vm.zone_reclaim_mode=0

# overcommit threshhold = 80%
vm.overcommit_memory=2
vm.overcommit_ratio=80

# vm.dirty_background_bytes=67108864 # 64MB mem (2xRAID cache) wake the bgwriter
vm.dirty_background_ratio=3       # latency-performance default
vm.dirty_ratio=10                 # latency-performance default

# deny access on 0x00000 - 0x10000
vm.mmap_min_addr=65536

#-------------------------------------------------------------#
#                        Filesystem                           #
#-------------------------------------------------------------#
# max open files: 382589 -> 167772160
fs.file-max=167772160

# max concurrent unfinished async io, should be larger than 1M.  65536->1M
fs.aio-max-nr=1048576


#-------------------------------------------------------------#
#                          Network                            #
#-------------------------------------------------------------#
# max connection in listen queue (triggers retrans if full)
net.core.somaxconn=65535
net.core.netdev_max_backlog=8192
# tcp receive/transmit buffer default = 256KiB
net.core.rmem_default=262144
net.core.wmem_default=262144
# receive/transmit buffer limit = 4MiB
net.core.rmem_max=4194304
net.core.wmem_max=4194304

# ip options
net.ipv4.ip_forward=1
net.ipv4.ip_nonlocal_bind=1
net.ipv4.ip_local_port_range=32768 65000

# tcp options
net.ipv4.tcp_timestamps=1
net.ipv4.tcp_tw_reuse=1
net.ipv4.tcp_tw_recycle=0
net.ipv4.tcp_syncookies=0
net.ipv4.tcp_synack_retries=1
net.ipv4.tcp_syn_retries=1

# tcp read/write buffer
net.ipv4.tcp_rmem="4096 87380 16777216"
net.ipv4.tcp_wmem="4096 16384 16777216"
net.ipv4.udp_mem="3145728 4194304 16777216"

# tcp probe fail interval: 75s -> 20s
net.ipv4.tcp_keepalive_intvl=20
# tcp break after 3 * 20s = 1m
net.ipv4.tcp_keepalive_probes=3
# probe peroid = 1 min
net.ipv4.tcp_keepalive_time=60

net.ipv4.tcp_fin_timeout=5
net.ipv4.tcp_max_tw_buckets=262144
net.ipv4.tcp_max_syn_backlog=8192
net.ipv4.neigh.default.gc_thresh1=80000
net.ipv4.neigh.default.gc_thresh2=90000
net.ipv4.neigh.default.gc_thresh3=100000

net.bridge.bridge-nf-call-iptables=1
net.bridge.bridge-nf-call-ip6tables=1
net.bridge.bridge-nf-call-arptables=1

# max connection tracking number
net.netfilter.nf_conntrack_max=1048576

2 - TINY

Tuned TINY模板

Tuned TINY模板主要针对极低配置的虚拟机进行优化,

此模板针对的典型机型是1核/1GB的虚拟机节点。您可以根据自己的实际机型进行调整。

# tuned configuration
#==============================================================#
# File      :   tuned.conf
# Mtime     :   2020-06-29
# Desc      :   Tune operatiing system to tiny mode
# Path      :   /etc/tuned/tiny/tuned.conf
# Author    :   Vonng(fengruohang@outlook.com)
# Copyright (C) 2018-2021 Ruohang Feng
#==============================================================#

[main]
summary=Optimize for PostgreSQL TINY System
# include=virtual-guest

[vm]
# disable transparent hugepages
transparent_hugepages=never

[sysctl]
#-------------------------------------------------------------#
#                           KERNEL                            #
#-------------------------------------------------------------#
# disable numa balancing
kernel.numa_balancing=0

# If a workload mostly uses anonymous memory and it hits this limit, the entire
# working set is buffered for I/O, and any more write buffering would require
# swapping, so it's time to throttle writes until I/O can catch up.  Workloads
# that mostly use file mappings may be able to use even higher values.
#
# The generator of dirty data starts writeback at this percentage (system default
# is 20%)
vm.dirty_ratio = 40

# Filesystem I/O is usually much more efficient than swapping, so try to keep
# swapping low.  It's usually safe to go even lower than this on systems with
# server-grade storage.
vm.swappiness = 30

#-------------------------------------------------------------#
#                          Network                            #
#-------------------------------------------------------------#
# tcp options
net.ipv4.tcp_timestamps=1
net.ipv4.tcp_tw_reuse=1
net.ipv4.tcp_tw_recycle=0
net.ipv4.tcp_syncookies=0
net.ipv4.tcp_synack_retries=1
net.ipv4.tcp_syn_retries=1

# tcp probe fail interval: 75s -> 20s
net.ipv4.tcp_keepalive_intvl=20
# tcp break after 3 * 20s = 1m
net.ipv4.tcp_keepalive_probes=3
# probe peroid = 1 min
net.ipv4.tcp_keepalive_time=60

3 - OLAP

Tuned OLAP模板,针对高并行,长查询,高吞吐实例优化

Tuned OLAP模板主要针对吞吐量与计算并行度进行优化

此模板针对的机型是Dell R740 64核/400GB内存,使用PCI-E SSD的节点。您可以根据自己的实际机型进行调整。

# tuned configuration
#==============================================================#
# File      :   tuned.conf
# Mtime     :   2020-09-18
# Desc      :   Tune operatiing system to olap mode
# Path      :   /etc/tuned/olap/tuned.conf
# Author    :   Vonng(fengruohang@outlook.com)
# Copyright (C) 2018-2021 Ruohang Feng
#==============================================================#

[main]
summary=Optimize for PostgreSQL OLAP System
include=network-throughput

[cpu]
force_latency=1
governor=performance
energy_perf_bias=performance
min_perf_pct=100

[vm]
# disable transparent hugepages
transparent_hugepages=never

[sysctl]
#-------------------------------------------------------------#
#                           KERNEL                            #
#-------------------------------------------------------------#
# disable numa balancing
kernel.numa_balancing=0

# total shmem size in bytes: $(expr $(getconf _PHYS_PAGES) / 2 \* $(getconf PAGE_SIZE))
{% if param_shmall is defined and param_shmall != '' %}
kernel.shmall = {{ param_shmall }}
{% endif %}

# total shmem size in pages:  $(expr $(getconf _PHYS_PAGES) / 2)
{% if param_shmmax is defined and param_shmmax != '' %}
kernel.shmmax = {{ param_shmmax }}
{% endif %}

# total shmem segs 4096 -> 8192
kernel.shmmni=8192

# total msg queue number, set to mem size in MB
kernel.msgmni=32768

# max length of message queue
kernel.msgmnb=65536

# max size of message
kernel.msgmax=65536

kernel.pid_max=131072

# max(Sem in Set)=2048, max(Sem)=max(Sem in Set) x max(SemSet) , max(Sem per Ops)=2048, max(SemSet)=65536
kernel.sem=2048 134217728 2048 65536

# do not sched postgres process in group
kernel.sched_autogroup_enabled = 0

# total time the scheduler will consider a migrated process cache hot and, thus, less likely to be remigrated
# defaut = 0.5ms (500000ns), update to 5ms , depending on your typical query (e.g < 1ms)
kernel.sched_migration_cost_ns=5000000

#-------------------------------------------------------------#
#                             VM                              #
#-------------------------------------------------------------#
# try not using swap
# vm.swappiness=10

# disable when most mem are for file cache
vm.zone_reclaim_mode=0

# overcommit threshhold = 80%
vm.overcommit_memory=2
vm.overcommit_ratio=80

vm.dirty_background_ratio = 10    # throughput-performance default
vm.dirty_ratio=80                 # throughput-performance default 40 -> 80

# deny access on 0x00000 - 0x10000
vm.mmap_min_addr=65536

#-------------------------------------------------------------#
#                        Filesystem                           #
#-------------------------------------------------------------#
# max open files: 382589 -> 167772160
fs.file-max=167772160

# max concurrent unfinished async io, should be larger than 1M.  65536->1M
fs.aio-max-nr=1048576


#-------------------------------------------------------------#
#                          Network                            #
#-------------------------------------------------------------#
# max connection in listen queue (triggers retrans if full)
net.core.somaxconn=65535
net.core.netdev_max_backlog=8192
# tcp receive/transmit buffer default = 256KiB
net.core.rmem_default=262144
net.core.wmem_default=262144
# receive/transmit buffer limit = 4MiB
net.core.rmem_max=4194304
net.core.wmem_max=4194304

# ip options
net.ipv4.ip_forward=1
net.ipv4.ip_nonlocal_bind=1
net.ipv4.ip_local_port_range=32768 65000

# tcp options
net.ipv4.tcp_timestamps=1
net.ipv4.tcp_tw_reuse=1
net.ipv4.tcp_tw_recycle=0
net.ipv4.tcp_syncookies=0
net.ipv4.tcp_synack_retries=1
net.ipv4.tcp_syn_retries=1

# tcp read/write buffer
net.ipv4.tcp_rmem="4096 87380 16777216"
net.ipv4.tcp_wmem="4096 16384 16777216"
net.ipv4.udp_mem="3145728 4194304 16777216"

# tcp probe fail interval: 75s -> 20s
net.ipv4.tcp_keepalive_intvl=20
# tcp break after 3 * 20s = 1m
net.ipv4.tcp_keepalive_probes=3
# probe peroid = 1 min
net.ipv4.tcp_keepalive_time=60

net.ipv4.tcp_fin_timeout=5
net.ipv4.tcp_max_tw_buckets=262144
net.ipv4.tcp_max_syn_backlog=8192
net.ipv4.neigh.default.gc_thresh1=80000
net.ipv4.neigh.default.gc_thresh2=90000
net.ipv4.neigh.default.gc_thresh3=100000

net.bridge.bridge-nf-call-iptables=1
net.bridge.bridge-nf-call-ip6tables=1
net.bridge.bridge-nf-call-arptables=1

# max connection tracking number
net.netfilter.nf_conntrack_max=1048576

4 - CRIT

Tuned CRIT模板,针对金融场景、不允许数据丢失错漏的场景进行优化。

Tuned CRIT模板主要针对RPO进行优化,尽可能减少内存中脏数据的量。

此模板针对的机型是Dell R740 64核/400GB内存,使用PCI-E SSD的节点。您可以根据自己的实际机型进行调整。

# tuned configuration
#==============================================================#
# File      :   tuned.conf
# Mtime     :   2020-06-29
# Desc      :   Tune operatiing system to crit mode
# Path      :   /etc/tuned/crit/tuned.conf
# Author    :   Vonng(fengruohang@outlook.com)
# Copyright (C) 2018-2021 Ruohang Feng
#==============================================================#

[main]
summary=Optimize for PostgreSQL CRIT System
include=network-latency

[cpu]
force_latency=1
governor=performance
energy_perf_bias=performance
min_perf_pct=100

[vm]
# disable transparent hugepages
transparent_hugepages=never

[sysctl]
#-------------------------------------------------------------#
#                           KERNEL                            #
#-------------------------------------------------------------#
# disable numa balancing
kernel.numa_balancing=0

# total shmem size in bytes: $(expr $(getconf _PHYS_PAGES) / 2 \* $(getconf PAGE_SIZE))
{% if param_shmall is defined and param_shmall != '' %}
kernel.shmall = {{ param_shmall }}
{% endif %}

# total shmem size in pages:  $(expr $(getconf _PHYS_PAGES) / 2)
{% if param_shmmax is defined and param_shmmax != '' %}
kernel.shmmax = {{ param_shmmax }}
{% endif %}

# total shmem segs 4096 -> 8192
kernel.shmmni=8192

# total msg queue number, set to mem size in MB
kernel.msgmni=32768

# max length of message queue
kernel.msgmnb=65536

# max size of message
kernel.msgmax=65536

kernel.pid_max=131072

# max(Sem in Set)=2048, max(Sem)=max(Sem in Set) x max(SemSet) , max(Sem per Ops)=2048, max(SemSet)=65536
kernel.sem=2048 134217728 2048 65536

# do not sched postgres process in group
kernel.sched_autogroup_enabled = 0

# total time the scheduler will consider a migrated process cache hot and, thus, less likely to be remigrated
# defaut = 0.5ms (500000ns), update to 5ms , depending on your typical query (e.g < 1ms)
kernel.sched_migration_cost_ns=5000000

#-------------------------------------------------------------#
#                             VM                              #
#-------------------------------------------------------------#
# try not using swap
vm.swappiness=0

# disable when most mem are for file cache
vm.zone_reclaim_mode=0

# overcommit threshhold = 80%
vm.overcommit_memory=2
vm.overcommit_ratio=100

# 64MB mem (2xRAID cache) wake the bgwriter
vm.dirty_background_bytes=67108864
# vm.dirty_background_ratio=3       # latency-performance default
vm.dirty_ratio=6                    # latency-performance default

# deny access on 0x00000 - 0x10000
vm.mmap_min_addr=65536

#-------------------------------------------------------------#
#                        Filesystem                           #
#-------------------------------------------------------------#
# max open files: 382589 -> 167772160
fs.file-max=167772160

# max concurrent unfinished async io, should be larger than 1M.  65536->1M
fs.aio-max-nr=1048576


#-------------------------------------------------------------#
#                          Network                            #
#-------------------------------------------------------------#
# max connection in listen queue (triggers retrans if full)
net.core.somaxconn=65535
net.core.netdev_max_backlog=8192
# tcp receive/transmit buffer default = 256KiB
net.core.rmem_default=262144
net.core.wmem_default=262144
# receive/transmit buffer limit = 4MiB
net.core.rmem_max=4194304
net.core.wmem_max=4194304

# ip options
net.ipv4.ip_forward=1
net.ipv4.ip_nonlocal_bind=1
net.ipv4.ip_local_port_range=32768 65000

# tcp options
net.ipv4.tcp_timestamps=1
net.ipv4.tcp_tw_reuse=1
net.ipv4.tcp_tw_recycle=0
net.ipv4.tcp_syncookies=0
net.ipv4.tcp_synack_retries=1
net.ipv4.tcp_syn_retries=1

# tcp read/write buffer
net.ipv4.tcp_rmem="4096 87380 16777216"
net.ipv4.tcp_wmem="4096 16384 16777216"
net.ipv4.udp_mem="3145728 4194304 16777216"

# tcp probe fail interval: 75s -> 20s
net.ipv4.tcp_keepalive_intvl=20
# tcp break after 3 * 20s = 1m
net.ipv4.tcp_keepalive_probes=3
# probe peroid = 1 min
net.ipv4.tcp_keepalive_time=60

net.ipv4.tcp_fin_timeout=5
net.ipv4.tcp_max_tw_buckets=262144
net.ipv4.tcp_max_syn_backlog=8192
net.ipv4.neigh.default.gc_thresh1=80000
net.ipv4.neigh.default.gc_thresh2=90000
net.ipv4.neigh.default.gc_thresh3=100000

net.bridge.bridge-nf-call-iptables=1
net.bridge.bridge-nf-call-ip6tables=1
net.bridge.bridge-nf-call-arptables=1

# max connection tracking number
net.netfilter.nf_conntrack_max=1048576