Introduction
Linux kernel tuning can dramatically improve application monitoring-practical-introduction" title="eBPF for Performance Monitoring: A Practical Introduction" class="internal-link">performance, but requires enterprise-computing" title="Understanding Quantum Supremacy: What It Means for Enterprise Computing" class="internal-link">understanding the workload characteristics and system behavior. This guide covers essential tuning parameters for high-performance enterprise environments.
Understanding Your Workload
Before tuning, characterize your workload:
- CPU-bound: Computation-intensive applications
- Memory-bound: Large working sets or memory bandwidth sensitive
- I/O-bound: Storage or network intensive
- Latency-sensitive: Real-time or interactive applications
CPU and Scheduler Tuning
CPU Governor
Set the performance governor for consistent speed:
# Check current governor
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
# Set performance governor
echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
NUMA Optimization
For multi-socket systems, optimize NUMA behavior:
# View NUMA topology
numactl --hardware
# Run process on specific NUMA node
numactl --cpunodebind=0 --membind=0 ./your_application
Scheduler Tuning
For latency-sensitive workloads:
# Reduce scheduler minimum granularity
echo 100000 > /proc/sys/kernel/sched_min_granularity_ns
# Reduce scheduler wakeup granularity
echo 500000 > /proc/sys/kernel/sched_wakeup_granularity_ns
Memory Management
Transparent Huge Pages
For most enterprise workloads, disable THP:
# Check current status
cat /sys/kernel/mm/transparent_hugepage/enabled
# Disable THP (recommended for databases)
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo never > /sys/kernel/mm/transparent_hugepage/defrag
Swappiness
Adjust swap behavior based on workload:
# Check current value
cat /proc/sys/vm/swappiness
# For servers with ample RAM
echo 10 > /proc/sys/vm/swappiness
# For databases (minimize swapping)
echo 1 > /proc/sys/vm/swappiness
Dirty Page Management
Tune write-back behavior for I/O workloads:
# Percentage of memory for dirty pages
echo 10 > /proc/sys/vm/dirty_ratio
echo 5 > /proc/sys/vm/dirty_background_ratio
# For high-throughput writes
echo 40 > /proc/sys/vm/dirty_ratio
echo 10 > /proc/sys/vm/dirty_background_ratio
Network Stack Tuning
Buffer Sizes
Increase network buffer sizes for throughput:
# TCP buffer sizes
echo "4096 87380 16777216" > /proc/sys/net/ipv4/tcp_rmem
echo "4096 65536 16777216" > /proc/sys/net/ipv4/tcp_wmem
# Core network buffers
echo 16777216 > /proc/sys/net/core/rmem_max
echo 16777216 > /proc/sys/net/core/wmem_max
Connection Handling
For high-connection servers:
# Increase connection backlog
echo 65535 > /proc/sys/net/core/somaxconn
echo 65535 > /proc/sys/net/ipv4/tcp_max_syn_backlog
# Enable TCP fast open
echo 3 > /proc/sys/net/ipv4/tcp_fastopen
# Reuse TIME_WAIT sockets
echo 1 > /proc/sys/net/ipv4/tcp_tw_reuse
I/O Subsystem
Scheduler Selection
Choose appropriate I/O scheduler:
# Check available schedulers
cat /sys/block/sda/queue/scheduler
# For SSDs, use none or mq-deadline
echo none > /sys/block/sda/queue/scheduler
# For HDDs, use mq-deadline
echo mq-deadline > /sys/block/sda/queue/scheduler
Read-Ahead
Tune read-ahead for sequential workloads:
# Check current value (in 512-byte sectors)
cat /sys/block/sda/queue/read_ahead_kb
# Increase for large sequential reads
echo 2048 > /sys/block/sda/queue/read_ahead_kb
Making Changes Permanent
Use sysctl.conf for persistent settings:
# /etc/sysctl.d/99-performance.conf
vm.swappiness = 10
vm.dirty_ratio = 10
vm.dirty_background_ratio = 5
net.core.somaxconn = 65535
net.ipv4.tcp_max_syn_backlog = 65535
Apply with: sysctl -p /etc/sysctl.d/99-performance.conf
Monitoring Impact
Always measure before and after tuning:
# System statistics
vmstat 1
iostat -x 1
mpstat -P ALL 1
# Detailed performance analysis
perf stat -d ./your_application
Conclusion
Kernel tuning is workload-specific. Start with understanding your application's behavior, make incremental changes, and always measure the impact. Document all changes and maintain configuration management for reproducibility.