Configure Linux transparent huge pages and memory optimization for database workloads

Check current THP status

First, examine your system's current transparent huge pages configuration to understand the baseline.

cat /sys/kernel/mm/transparent_hugepage/enabled
cat /sys/kernel/mm/transparent_hugepage/defrag
grep -i hugepages /proc/meminfo

Monitor THP allocation statistics

View detailed statistics about huge page allocation and fragmentation events.

cat /proc/vmstat | grep -E "thp_|hugepages"
cat /sys/kernel/mm/transparent_hugepage/khugepaged/pages_collapsed
cat /sys/kernel/mm/transparent_hugepage/khugepaged/full_scans

Create THP management script

Create a comprehensive script to manage transparent huge pages settings dynamically.

sudo mkdir -p /usr/local/bin
sudo tee /usr/local/bin/thp-manager.sh

#!/bin/bash

# THP Manager Script for Database Workloads
# Usage: thp-manager.sh [disable|enable|madvise|status]

THP_ENABLED="/sys/kernel/mm/transparent_hugepage/enabled"
THP_DEFRAG="/sys/kernel/mm/transparent_hugepage/defrag"
THP_KHUGEPAGED="/sys/kernel/mm/transparent_hugepage/khugepaged"

set_thp_status() {
    local mode=$1
    case $mode in
        "disable")
            echo "never" > $THP_ENABLED
            echo "never" > $THP_DEFRAG
            echo "THP disabled completely"
            ;;
        "enable")
            echo "always" > $THP_ENABLED
            echo "always" > $THP_DEFRAG
            echo "THP enabled with aggressive allocation"
            ;;
        "madvise")
            echo "madvise" > $THP_ENABLED
            echo "madvise" > $THP_DEFRAG
            echo "THP set to application-controlled mode"
            ;;
    esac
}

show_status() {
    echo "=== Transparent Huge Pages Status ==="
    echo "Enabled: $(cat $THP_ENABLED)"
    echo "Defrag: $(cat $THP_DEFRAG)"
    echo "khugepaged scan pages: $(cat $THP_KHUGEPAGED/pages_to_scan)"
    echo "khugepaged scan sleep: $(cat $THP_KHUGEPAGED/scan_sleep_millisecs)ms"
    echo
    echo "=== Memory Statistics ==="
    grep -E "HugePages_|AnonHugePages" /proc/meminfo
    echo
    echo "=== THP Events ==="
    grep -E "thp_" /proc/vmstat | head -10
}

optimize_khugepaged() {
    # Optimize khugepaged for database workloads
    echo 4096 > $THP_KHUGEPAGED/pages_to_scan
    echo 1000 > $THP_KHUGEPAGED/scan_sleep_millisecs
    echo 32 > $THP_KHUGEPAGED/alloc_sleep_millisecs
    echo "khugepaged optimized for database workloads"
}

case "$1" in
    disable)
        set_thp_status "disable"
        ;;
    enable)
        set_thp_status "enable"
        optimize_khugepaged
        ;;
    madvise)
        set_thp_status "madvise"
        optimize_khugepaged
        ;;
    status)
        show_status
        ;;
    *)
        echo "Usage: $0 {disable|enable|madvise|status}"
        echo "  disable  - Completely disable THP (recommended for most databases)"
        echo "  enable   - Enable THP with aggressive allocation"
        echo "  madvise  - Enable THP only when requested by applications"
        echo "  status   - Show current THP configuration and statistics"
        exit 1
        ;;
esac

sudo chmod +x /usr/local/bin/thp-manager.sh

Configure THP for database workloads

Most database systems perform better with THP disabled due to memory fragmentation issues.

# Check current status
sudo /usr/local/bin/thp-manager.sh status

# Disable THP for database workloads (recommended)
sudo /usr/local/bin/thp-manager.sh disable

# Verify the change
sudo /usr/local/bin/thp-manager.sh status

Create persistent THP configuration

Make THP settings persistent across reboots using kernel parameters and systemd service.

# Update GRUB configuration
sudo tee -a /etc/default/grub.d/thp.cfg

# Disable Transparent Huge Pages for database workloads
GRUB_CMDLINE_LINUX_DEFAULT="$GRUB_CMDLINE_LINUX_DEFAULT transparent_hugepage=never"

sudo update-grub

# Update GRUB configuration
sudo tee -a /etc/default/grub.d/thp.cfg

# Disable Transparent Huge Pages for database workloads
GRUB_CMDLINE_LINUX_DEFAULT="$GRUB_CMDLINE_LINUX_DEFAULT transparent_hugepage=never"

sudo grub2-mkconfig -o /boot/grub2/grub.cfg

Create systemd service for THP management

Create a systemd service to apply THP settings early in the boot process as a backup to kernel parameters.

sudo tee /etc/systemd/system/thp-config.service

[Unit]
Description=Configure Transparent Huge Pages for Database Workloads
DefaultDependencies=no
After=sysinit.target local-fs.target
Before=basic.target

[Service]
Type=oneshot
ExecStart=/usr/local/bin/thp-manager.sh disable
TimeoutSec=30
RemainAfterExit=yes

[Install]
WantedBy=basic.target

sudo systemctl enable thp-config.service
sudo systemctl daemon-reload

Configure memory allocation for PostgreSQL

Optimize PostgreSQL memory settings when THP is disabled to prevent performance degradation.

sudo tee /etc/sysctl.d/99-postgresql-memory.conf

# PostgreSQL memory optimization with THP disabled

# Reduce memory fragmentation
vm.swappiness = 10

# Optimize dirty page handling
vm.dirty_ratio = 15
vm.dirty_background_ratio = 5
vm.dirty_expire_centisecs = 3000
vm.dirty_writeback_centisecs = 500

# Memory allocation optimization
vm.overcommit_memory = 2
vm.overcommit_ratio = 80

# Shared memory settings
kernel.shmmax = 17179869184
kernel.shmall = 4194304
kernel.shmmni = 4096

sudo sysctl --load=/etc/sysctl.d/99-postgresql-memory.conf

Configure memory settings for Redis

Redis requires specific memory optimizations when running without THP support.

sudo tee /etc/sysctl.d/99-redis-memory.conf

# Redis memory optimization with THP disabled

# Enable memory overcommit (required for Redis background saves)
vm.overcommit_memory = 1

# Optimize for Redis memory patterns
vm.swappiness = 1
vm.vfs_cache_pressure = 50

# Network buffer optimization for Redis replication
net.core.somaxconn = 65535
net.core.tcp_max_syn_backlog = 65535

sudo sysctl --load=/etc/sysctl.d/99-redis-memory.conf

Configure MongoDB memory optimization

MongoDB has specific requirements when THP is disabled to maintain performance.

sudo tee /etc/sysctl.d/99-mongodb-memory.conf

# MongoDB memory optimization with THP disabled

# Optimize for MongoDB's memory access patterns
vm.swappiness = 1
vm.dirty_ratio = 15
vm.dirty_background_ratio = 5

# Reduce zone reclaim for NUMA systems
vm.zone_reclaim_mode = 0

# Network optimization for MongoDB clusters
net.core.somaxconn = 4096
net.ipv4.tcp_keepalive_time = 300

sudo sysctl --load=/etc/sysctl.d/99-mongodb-memory.conf

Create THP monitoring script

Implement comprehensive monitoring to track THP impact on database performance.

sudo tee /usr/local/bin/thp-monitor.sh

#!/bin/bash

# THP Performance Monitor
# Collects THP statistics and memory fragmentation metrics

TIMESTAMP=$(date '+%Y-%m-%d %H:%M:%S')
LOGFILE="/var/log/thp-monitor.log"

log_metrics() {
    echo "=== THP Metrics - $TIMESTAMP ===" >> $LOGFILE
    
    # Basic THP status
    echo "THP_ENABLED: $(cat /sys/kernel/mm/transparent_hugepage/enabled)" >> $LOGFILE
    echo "THP_DEFRAG: $(cat /sys/kernel/mm/transparent_hugepage/defrag)" >> $LOGFILE
    
    # Memory statistics
    grep -E "AnonHugePages|HugePages_" /proc/meminfo >> $LOGFILE
    
    # THP allocation statistics
    grep -E "thp_fault_alloc|thp_fault_fallback|thp_collapse_alloc" /proc/vmstat >> $LOGFILE
    
    # Memory fragmentation
    echo "FRAGMENTATION_INDEX:" >> $LOGFILE
    cat /sys/kernel/debug/extfrag/extfrag_index 2>/dev/null | head -5 >> $LOGFILE
    
    # Buddy allocator info
    echo "BUDDY_INFO:" >> $LOGFILE
    cat /proc/buddyinfo >> $LOGFILE
    
    echo "" >> $LOGFILE
}

generate_report() {
    echo "=== THP Performance Report - $TIMESTAMP ==="
    
    # Calculate THP efficiency
    FAULT_ALLOC=$(grep thp_fault_alloc /proc/vmstat | awk '{print $2}')
    FAULT_FALLBACK=$(grep thp_fault_fallback /proc/vmstat | awk '{print $2}')
    
    if [ "$FAULT_ALLOC" -gt 0 ] && [ "$FAULT_FALLBACK" -gt 0 ]; then
        TOTAL_ATTEMPTS=$((FAULT_ALLOC + FAULT_FALLBACK))
        SUCCESS_RATE=$((FAULT_ALLOC * 100 / TOTAL_ATTEMPTS))
        echo "THP Success Rate: $SUCCESS_RATE%"
    fi
    
    # Memory pressure indicators
    DIRECT_RECLAIM=$(grep pgsteal_direct /proc/vmstat | awk '{print $2}')
    KSWAPD_RECLAIM=$(grep pgsteal_kswapd /proc/vmstat | awk '{print $2}')
    echo "Direct Reclaim Events: $DIRECT_RECLAIM"
    echo "kswapd Reclaim Events: $KSWAPD_RECLAIM"
    
    # Current memory usage
    echo ""
    echo "Current Memory Usage:"
    free -h
    
    echo ""
    echo "Memory Fragmentation (lower is better):"
    awk '/Node 0/ && /DMA32|Normal/ {print $1 " " $2 ": " $NF}' /proc/buddyinfo 2>/dev/null || echo "Fragmentation data not available"
}

case "$1" in
    log)
        log_metrics
        ;;
    report)
        generate_report
        ;;
    watch)
        watch -n 5 "$0 report"
        ;;
    *)
        echo "Usage: $0 {log|report|watch}"
        echo "  log    - Log current metrics to $LOGFILE"
        echo "  report - Generate performance report"
        echo "  watch  - Continuously monitor (5-second intervals)"
        exit 1
        ;;
esac

sudo chmod +x /usr/local/bin/thp-monitor.sh
sudo touch /var/log/thp-monitor.log

Set up automated THP monitoring

Configure systemd timer to regularly collect THP performance metrics.

sudo tee /etc/systemd/system/thp-monitor.service

[Unit]
Description=THP Performance Monitoring
After=multi-user.target

[Service]
Type=oneshot
User=root
ExecStart=/usr/local/bin/thp-monitor.sh log
StandardOutput=journal
StandardError=journal

sudo tee /etc/systemd/system/thp-monitor.timer

[Unit]
Description=THP Performance Monitoring Timer
Requires=thp-monitor.service

[Timer]
OnBootSec=5min
OnUnitActiveSec=10min
Persistent=true

[Install]
WantedBy=timers.target

sudo systemctl enable thp-monitor.timer
sudo systemctl start thp-monitor.timer
sudo systemctl daemon-reload

Configure logrotate for THP monitoring

Prevent THP monitoring logs from consuming excessive disk space.

sudo tee /etc/logrotate.d/thp-monitor

/var/log/thp-monitor.log {
    daily
    rotate 30
    compress
    delaycompress
    missingok
    notifempty
    create 644 root root
    postrotate
        systemctl reload rsyslog > /dev/null 2>&1 || true
    endscript
}

Configure NUMA memory policy

On NUMA systems, configure memory policies to optimize database performance with disabled THP.

# Check NUMA topology
numactl --hardware

# Configure NUMA-aware memory allocation
sudo tee /etc/sysctl.d/99-numa-memory.conf

# NUMA memory optimization for databases

# Disable zone reclaim to prevent remote memory access penalties
vm.zone_reclaim_mode = 0

# Optimize NUMA balancing for database workloads
kernel.numa_balancing = 0

# Memory allocation preferences
vm.numa_zonelist_order = Node

sudo sysctl --load=/etc/sysctl.d/99-numa-memory.conf

Implement memory pressure monitoring

Set up monitoring for memory pressure indicators that can impact database performance.

sudo tee /usr/local/bin/memory-pressure-check.sh

#!/bin/bash

# Memory Pressure Monitoring for Database Systems

WARN_THRESHOLD=80
CRIT_THRESHOLD=95

check_memory_pressure() {
    # Get memory usage percentage
    MEMORY_USAGE=$(free | awk 'NR==2{printf "%.0f", $3*100/$2}')
    
    # Check swap usage
    SWAP_USAGE=$(free | awk 'NR==3{if($2>0) printf "%.0f", $3*100/$2; else print "0"}')
    
    # Get memory pressure stall information (if available)
    if [ -f /proc/pressure/memory ]; then
        PRESSURE_AVG10=$(awk '/some/ {print $2}' /proc/pressure/memory | cut -d'=' -f2)
        PRESSURE_TOTAL=$(awk '/some/ {print $4}' /proc/pressure/memory | cut -d'=' -f2)
    fi
    
    # Check for memory fragmentation
    FRAGMENTATION=$(awk '/Node 0.*Normal/ {print $(NF-1)}' /proc/buddyinfo 2>/dev/null | head -1)
    
    echo "Memory Usage: ${MEMORY_USAGE}%"
    echo "Swap Usage: ${SWAP_USAGE}%"
    
    if [ ! -z "$PRESSURE_AVG10" ]; then
        echo "Memory Pressure (10s avg): ${PRESSURE_AVG10}%"
    fi
    
    # Alert conditions
    if [ "$MEMORY_USAGE" -ge "$CRIT_THRESHOLD" ]; then
        echo "CRITICAL: Memory usage above ${CRIT_THRESHOLD}%"
        logger "CRITICAL: High memory usage detected: ${MEMORY_USAGE}%"
    elif [ "$MEMORY_USAGE" -ge "$WARN_THRESHOLD" ]; then
        echo "WARNING: Memory usage above ${WARN_THRESHOLD}%"
        logger "WARNING: Memory usage elevated: ${MEMORY_USAGE}%"
    fi
    
    if [ "$SWAP_USAGE" -gt "10" ]; then
        echo "WARNING: Swap usage detected: ${SWAP_USAGE}%"
        logger "WARNING: System using swap memory: ${SWAP_USAGE}%"
    fi
}

check_memory_pressure

sudo chmod +x /usr/local/bin/memory-pressure-check.sh

Verify THP configuration

Confirm that transparent huge pages are properly configured for your database workload.

# Check THP status
sudo /usr/local/bin/thp-manager.sh status

# Verify kernel parameters will persist after reboot
grep transparent_hugepage /proc/cmdline

# Check systemd service status
sudo systemctl status thp-config.service

Monitor performance impact

Generate a performance report and verify monitoring is working correctly.

# Generate current performance report
sudo /usr/local/bin/thp-monitor.sh report

# Check monitoring timer status
sudo systemctl status thp-monitor.timer

# View recent monitoring data
sudo tail -20 /var/log/thp-monitor.log

# Test memory pressure monitoring
sudo /usr/local/bin/memory-pressure-check.sh

Test database performance

Run basic performance tests to verify the THP configuration benefits your database workload. This setup improves database performance by reducing memory fragmentation and eliminating THP-related latency spikes that can impact query response times.

# Check system memory allocation efficiency
grep -E "thp_fault_fallback|thp_fault_alloc" /proc/vmstat

# Monitor memory fragmentation
cat /proc/buddyinfo | awk '{print $1, $2, $NF}'

# Check for memory reclaim events (should be minimal)
grep -E "pgsteal_|pgscan_" /proc/vmstat