Skip to content

Memory Tuning

Linux kernel memory management parameters allow fine-grained control over how the system allocates, swaps, and caches memory. Tuning these parameters can improve performance for workloads like databases, virtualization, or high-throughput applications. Key parameters include vm.swappiness, kernel.shmall, and others that balance memory usage between application needs and system stability.

vm.swappiness: Controlling Swapping Behavior

The vm.swappiness parameter determines how aggressively the kernel will swap memory pages to disk. A higher value (0–100) encourages swapping, while a lower value reduces it.

Default and Behavior

  • Default: 60 (balanced approach)
  • 0: Minimizes swapping (suitable for servers with ample RAM)
  • 100: Aggressively swaps (useful for systems with limited RAM)

Example:

# Check current value  
sysctl vm.swappiness  

# Temporarily set to 10 (reduces swapping)  
sudo sysctl -w vm.swappiness=10  

# Persistently apply (edit /etc/sysctl.conf)  
echo "vm.swappiness=10" | sudo tee -a /etc/sysctl.conf

kernel.shmall and kernel.shmmax: Shared Memory Limits

These parameters define the maximum size of shared memory segments (in pages) and the maximum size of a single segment, respectively. They are critical for applications like databases or message queues that rely on shared memory.

Calculation and Defaults

  • kernel.shmall (pages): Default is typically 2097152 (2GB on 4KB pages)
  • kernel.shmmax (bytes): Default is often 134217728 (128MB)

Example:

# Check current values  
sysctl kernel.shmall  
sysctl kernel.shmmax  

# Set new values (e.g., 4GB shared memory)  
sudo sysctl -w kernel.shmall=524288  
sudo sysctl -w kernel.shmmax=4294967296

Note: Ensure kernel.shmmax does not exceed the system’s physical memory (RAM + swap).

vm.vfs_cache_pressure: File System Caching

This parameter controls how aggressively the kernel evicts file system cache to free memory. Higher values reduce caching, which may improve responsiveness for memory-intensive tasks.

  • Default: 100
  • Values > 100: Prioritize application memory over caching
  • Values < 10: Prioritize caching

Example:

# Set to 150 (reduce caching pressure)  
sudo sysctl -w vm.vfs_cache_pressure=150

vm.overcommit_memory and vm.overcommit_ratio: Memory Overcommit

These parameters control how the kernel manages memory allocation for processes.

  • vm.overcommit_memory=0: Heuristics-based overcommit (default)
  • vm.overcommit_memory=1: Always overcommit (useful for applications like databases)
  • vm.overcommit_ratio: Percentage of RAM allowed for overcommit (default: 50%)

Example:

# Enable strict overcommit for a database server  
sudo sysctl -w vm.overcommit_memory=1  
sudo sysctl -w vm.overcommit_ratio=75

vm.min_free_kbytes: Minimum Free Memory

This parameter sets the minimum amount of memory (in KB) that must always be free. It prevents the system from using all available memory, which can cause instability.

  • Default: ~16384 KB (16MB) for most systems
  • Adjust based on workload: Higher values for critical services, lower for high-memory systems

Example:

# Set minimum free memory to 32MB  
sudo sysctl -w vm.min_free_kbytes=32768

Key takeaways

  • vm.swappiness balances swapping behavior; set to 10 for servers.
  • kernel.shmall and kernel.shmmax define shared memory limits; ensure they align with application needs.
  • Adjust vm.vfs_cache_pressure to prioritize memory for critical workloads.
  • Use vm.overcommit_memory and vm.overcommit_ratio to optimize memory allocation for specific use cases.
  • Always test changes in non-production environments and validate with tools like free, top, and vmstat.