How to Simulate Slow Disk I/O for MySQL Experiments

The article addresses the problem of reproducing slow disk I/O conditions to study MySQL behavior when its binlog resides on a sluggish storage device.

Step 1 – Create a delayed disk: Use dd if=/dev/zero of=slow.img bs=1M count=100 to generate a 100 MiB file, attach it as a loop device with losetup /dev/loop3 slow.img, and then map the loop device to a dm‑target that adds a 100 ms latency to both read and write operations.

Step 2 – Prepare MySQL on the slow device: Format the delayed device, mount it, and install MySQL via dbdeployer. Configure the binary log directory to point to the mounted slow device (e.g., /mnt/slow) and enable double‑write flush parameters.

Run a workload with mysqlslap to generate I/O pressure. Monitoring with iostat shows that the dm‑0 device (the delayed mapper) experiences queueing and high utilization, while the underlying /dev/loop3 remains under‑utilized, confirming the introduced latency.

Further profiling with pt‑ioprofile reveals that the majority of MySQL’s I/O time is spent on binlog operations, highlighting the impact of the artificial slowdown.

Result: By creating a slow I/O device and placing MySQL’s binlog on it, the experiment provides a reproducible environment for analyzing MySQL’s behavior under local file I/O bottlenecks, enabling deeper performance investigations.