Tuning Dirty Page Expiry for High-Frequency Theme Logging

Environment: Debian 12 (Bookworm), 4 vCPUs, 8GB RAM, Ext4 filesystem on NVMe.

I noted a persistent 3% iowait on a virtual machine running the Fionca – Business Finance WordPress Theme. This occurred on an idle node with a 0.1 load average. Standard application logs showed nothing unusual. The CPU was spending cycles waiting for the disk, yet no significant traffic was present. This indicated a kernel-level writeback issue rather than a capacity bottleneck.

iotop -oPa was used to monitor the cumulative disk I/O. The output highlighted that jbd2/sda1-8, the ext4 journaling daemon, was waking up every five seconds to commit data. Correlating this with the PHP-FPM worker PIDs, I identified that the Fionca theme framework was writing to a local telemetry file. Specifically, the theme's internal financial calculator script logs every slider movement to a flat file. While comparing this behavior to other items in the Download WooCommerce Theme repository, it became clear that this specific theme lacks a write-buffer for its internal analytics logging.

The Linux kernel manages these writes through the page cache. By default, vm.dirty_background_ratio is set to 10%, meaning background writeback starts when dirty pages reach 10% of total memory. However, the dirty_expire_centisecs parameter dictates that pages older than 30 seconds must be flushed regardless of the ratio. The real bottleneck was the ext4 filesystem's commit mount option. By default, ext4 forces a journal commit every 5 seconds. Every single AJAX hit from the Fionca calculator was marking an inode as dirty, and the journal was flushing this metadata to the NVMe, causing the observed IOWait.

I investigated the VFS layer. Each write call from PHP-FPM requires an update to the inode's metadata, specifically the mtime. Since Fionca does not use asynchronous I/O for its logging, the PHP process waits for the kernel to acknowledge the write. To mitigate this, I adjusted the mount options to commit=60 and tuned the kernel sysctl parameters to increase the dirty page age. This allows the kernel to coalesce dozens of log entries in RAM before a single sequential write to the physical disk.

I also examined the theme's functions.php. The logging function used file_put_contents with the FILE_APPEND flag. In a Linux environment, this translates to an open() call with O_APPEND. By modifying the theme to use a Redis-based buffer for these calculator events, the disk I/O was eliminated. However, for those restricted to filesystem logging, kernel tuning is the only path. The result was a drop in iowait to 0.01%, stabilizing the TTFB for dynamic requests. Do not trust default mount options for sites with high-frequency telemetry scripts. Monitor /proc/meminfo to see Dirty page accumulation.

# /etc/sysctl.conf
vm.dirty_background_ratio = 5
vm.dirty_ratio = 10
vm.dirty_expire_centisecs = 6000
vm.dirty_writeback_centisecs = 1000

Increase the commit interval in /etc/fstab by adding commit=60 to the mount options.