Apache Worker MPM Performance Optimization: Threads vs Processes Tuning Guide


2 views

The Worker MPM in Apache 2.2 implements a hybrid multi-process/multi-threaded model. The key directives in httpd.conf are:

<IfModule mpm_worker_module>
    StartServers          2
    MaxClients          150
    MinSpareThreads      25
    MaxSpareThreads      75
    ThreadsPerChild      25
    MaxRequestsPerChild   0
</IfModule>

More processes advantages:

  • Better isolation - crashing threads won't bring down entire server
  • Better for CPU-bound workloads (avoids GIL contention in some cases)
  • Easier to scale vertically on multi-core systems

More threads advantages:

  • Lower memory overhead (shared address space)
  • Faster context switching between requests
  • Better for I/O-bound workloads handling many concurrent connections

For a database-heavy application (I/O bound):

<IfModule mpm_worker_module>
    ServerLimit          16
    ThreadLimit          64
    StartServers         4
    MaxClients          800  # 16 processes * 50 threads
    ThreadsPerChild      50
    MinSpareThreads     50
    MaxSpareThreads    200
</IfModule>

For CPU-intensive processing:

<IfModule mpm_worker_module>
    ServerLimit          32
    ThreadLimit          25  
    StartServers         8
    MaxClients          800  # 32 processes * 25 threads
    ThreadsPerChild      25
    MinSpareThreads     50
    MaxSpareThreads    150
</IfModule>

Always test with your specific workload using tools like ab (Apache Bench):

ab -n 1000 -c 100 http://yourserver/test.php

Monitor memory usage per process:

ps -ylC httpd --sort:rss

High CPU usage: Reduce ThreadsPerChild and increase ServerLimit

Memory exhaustion: Reduce total threads (MaxClients = ServerLimit × ThreadsPerChild)

Connection timeouts: Verify MaxRequestsPerChild isn't causing frequent restarts

The Worker MPM in Apache 2.2 uses a hybrid approach combining processes and threads. The key configuration directives are:

<IfModule mpm_worker_module>
    ServerLimit          16
    StartServers         4
    MaxClients          400
    MinSpareThreads      25
    MaxSpareThreads      75
    ThreadsPerChild      25
    MaxRequestsPerChild  0
</IfModule>

More processes advantages:

  • Better isolation - crashes affect fewer connections
  • Easier to debug with traditional tools
  • Better for environments with thread-unsafe libraries

More threads advantages:

  • Lower memory overhead (shared address space)
  • Faster context switching
  • Better for CPU-bound workloads

Memory-constrained environment:

# Optimize for memory efficiency
ThreadsPerChild 50
MaxClients 500
ServerLimit 10

High-traffic dynamic content:

# Balance isolation and performance
ThreadsPerChild 25
MaxClients 400
ServerLimit 16

Test configuration comparing different setups:

ab -n 10000 -c 100 http://localhost/test.php

# Configuration A (more processes)
ServerLimit 20
ThreadsPerChild 10
MaxClients 200

# Configuration B (more threads) 
ServerLimit 10
ThreadsPerChild 25
MaxClients 250

Memory leaks: Use lower ThreadsPerChild when dealing with leaky modules

Stability problems: Increase processes if experiencing random crashes

Connection limits: Calculate MaxClients based on available RAM

MaxClients = (Total RAM - OS overhead) / (Memory per process)

For newer Apache versions, consider:

  • Event MPM for better async handling
  • PHP-FPM for separating web server from application processing