Ubuntu Kernel Performance Tuning for High-Load Servers

Kernel tuning should start from workload telemetry, not internet preset files. The right values for a TLS-heavy API server differ from a queue consumer host, so baseline first and tune with evidence.

1. Measure before changing sysctl

vmstat 1 30
mpstat -P ALL 1 30
pidstat -dru 1 30
ss -s

2. Network and socket baseline

# /etc/sysctl.d/99-net-tuning.conf
net.core.somaxconn = 4096
net.core.netdev_max_backlog = 8192
net.ipv4.tcp_max_syn_backlog = 8192
net.ipv4.ip_local_port_range = 1024 65535
net.ipv4.tcp_fin_timeout = 15

3. Memory and VM pressure controls

# /etc/sysctl.d/99-vm-tuning.conf
vm.swappiness = 10
vm.dirty_background_ratio = 5
vm.dirty_ratio = 20
vm.max_map_count = 262144

Pair kernel tuning with process memory limits and OOM strategy. Kernel changes alone cannot protect a host from unbounded application memory growth.

4. Apply and validate safely

sudo sysctl --system
sudo systemd-analyze verify
sudo journalctl -p err -b | tail -n 50

5. Day-2 checklist

• Version-control every sysctl change with workload notes.
• Re-test after kernel upgrades and cloud image refreshes.
• Alert on sustained steal time, major page faults, and run queue spikes.
• Keep rollback profiles for each server class.

"Effective kernel tuning is iterative engineering: measure, change one variable, validate, and document."