Which MPM best serves mixed static/dynamic workloads in Apache 2.4?

For a server serving mixed static and dynamic content under Apache 2.4—especially with a high KeepAliveTimeout—mpm_event is the superior choice over mpm_worker.

The primary reason is management of idle connections. mpm_worker assigns a dedicated worker thread to the entire duration of a Keep-Alive session. If your timeout is set to 5 seconds, those threads remain tied up by clients not actively sending data, leading to thread pool exhaustion during traffic bursts. In contrast, mpm_event uses a dedicated listener thread to monitor Keep-Alive connections, only assigning a worker thread when there is an actual request to process.

Why Event Wins on Mixed Workloads

While both MPMs use a multi-threaded architecture, their handling of concurrency differs in ways that impact your specific constraints:

• Resource Efficiency: mpm_event frees up worker threads during the I/O wait times of Keep-Alive sessions. This allows the server to handle more concurrent static requests without hitting the thread limit.
• Latency: Under bursty traffic, mpm_event maintains lower average latency because dynamic requests are less likely to be queued waiting for a thread held by an idle static session.
• Memory Footprint: Because it doesn't maintain a thread for every idle connection, mpm_event typically consumes less resident set size (RSS) as connection counts scale.

Verification Methodology

To confirm this behavior on your specific hardware, follow this diagnostic path:

1. Enable Diagnostics: Ensure mod_status is enabled and set ExtendedStatus On in your Apache configuration.
2. Simulate Load: Use a load testing tool like wrk or ab (Apache Benchmark) to simulate bursty traffic with long-lived connections. For example:

   ab -n 10000 -c 500 -k http://your-server/

   (the -k flag enables Keep-Alive).
3. Monitor Thread State: Observe the server-status page. Look at the "Busy" vs. "Idle" thread counts. In mpm_worker, you will likely see Busy threads hit your MaxRequestWorkers limit quickly. In mpm_event, Busy threads should remain proportional to active requests only.
4. Measure Memory: Use top or htop to monitor the RSS of the Apache child processes during the load test to compare between the two modules.

Note: While mod_deflate increases CPU usage on both MPMs, mpm_event mitigates the impact on throughput by ensuring CPU cycles aren't wasted on managing blocked threads.